LatexMan.txt

\documentclass[12pt]{article}
\usepackage{amsmath}
\usepackage{amssymb}
\usepackage{epsfig}
\usepackage{graphicx}
\usepackage{color}
\usepackage{url}
\usepackage{caption}
\usepackage{epstopdf}


\begin{document}
%\bibliographystyle{plain}


\title{User Manual for SAHM package for Vistrails}
\author{Colin B. Talbert and Marian K. Talbert}
\maketitle
\vspace{2in}
\pagebreak

\tableofcontents

\pagebreak


\begin{flushleft}
\LARGE
\textbf{User Manual For for SAHM package for Vistrails} \\*

\normalsize
\vspace{5mm}
Colin B. Talbert and Marian K. Talbert
\vspace{1cm}
\end{flushleft}

\setlength{\parskip}{.5cm} 

\section{FieldData}


        The FieldData module allows a user to add presence/absence points or count data recorded
    across a landscape for the phenomenon being modeled (e.g., plant sightings, evidence of animal
    presence, etc.).  The input data for this module must be in the form of a .csv file that follows
    one of two formats:

        Format 1:
        A .csv file with the following column headings, in order: ``X," ``Y," and ``responseBinary".
    In this case, the ``X" field should be populated with the horizontal (longitudinal) positional
    data for a sample point. The ``Y" field should be populated with the vertical (latitudinal) data
    for a sample point. These values must be in the same coordinate system/units as the template
    layer used in the workflow. The column ``responseBinary" should be populated with either a `0'
    (indicating absence at the point) or a `1' (indicating presence at the point).

        Format 2:
        A .csv file with the following column headings, in order: ``X," ``Y," and ``responseCount".  In
    this case, the ``X" field should be populated with the horizontal (longitudinal) positional data
    for a sample point. The ``Y" field should be populated with the vertical (latitudinal) data for a
    sample point. These values must be in the same coordinate system/units as the template layer
    used in the workflow. The column ``responseCount" should be populated with either a `-9999'
    (indicating that the point is a background point) or a numerical value (either `0' or a positive
    integer) indicating the number of incidences of the phenomenon recorded at that point.

 \subsection*{Output Ports}
\begin{itemize}
\item value (mandatory)
        This is the actual file object that is being passed to other modules in the workflow.

\textbf{Common connections}

\begin{itemize}
\item The `fieldData\_file' input port of the FieldDataQuery Module if the field data needs subsetting or
aggregation.
\item The `fieldData' input port of the FieldDataAggregateAndWeight Module if the field data needs to be
aggregated or weighted to match the spatial resolution of the template layer.
\item The `fieldData' input port of the MDS builder Module if the field data needs no further pre-
processing prior to modeling.
\end{itemize}

\item value\_as\_string (optional)
        This is a VisTrails port that is not used in general SAHM workflows.

\textbf{Common connections}

\begin{itemize}
\item This does not commonly connect to other SAHM modules.
\end{itemize}

\end{itemize}


\section{Predictor}


        The Predictor module allows a user to select a single raster layer for consideration in the
    modeled analysis. Besides selecting the file the user also specifies the parameters to use for
    resampling, aggregation, and whether the data is categorical.

\subsection*{Input Ports}
\begin{itemize}
\item categorical (optional)
        This paramater allows a user to indicate the type of data represented.  The distinction
        between continuous and categorical data will maintained throught a workflow by appending the
        word `\_categorical' to categorical layer names in the resulting MDS file.  It is also import
        to select the nearest neighbor resampling option for categorical layers.

\textbf{Default value} = False (Unchecked)

\textbf{Options}

\begin{itemize}
\item True (Checked) - The data contained in the raster layer is categorical (e.g., landcover categories).
\item False(Unchecked) - The data contained in the raster is continuous (e.g., a DEM layer).
\end{itemize}

\item ResampleMethod (mandatory)
        The resample method employed to interpolate new cell values when transforming the raster
        layer to the coordinate space or cell size of the template layer.
\textbf{Options}

\begin{itemize}
\item near:  nearest neighbour resampling Fastest algorithm, worst interpolation quality, but best choice
for categorical data.
\item bilinear:  bilinear resampling, good choice for continuous data.
\item cubic:   cubic resampling.
\item cubicspline:  cubic spline resampling.
\item lanczos:  Lanczos windowed sinc resampling.
\item see: http://www.gdal.org/gdalwarp.html for context
\end{itemize}

\item AggregationMethod (mandatory)
        The aggregation method to be used in the event that the raster layer must be up-scaled to
        match the template layer (e.g., generalizing a 10 m input layer to a 100 m output layer).
        Care should be taken to ensure that the aggregation method that best preserves the integrity
        of the data is used.  See the PARC module documentation for more information on how
        resampling and aggregation are performed.
\textbf{Options}

\begin{itemize}
\item Mean:  Average value of all constituent pixels used.
\item Max:   Maximum value of all constituent pixels used.
\item Min:   Minimum value of all constituent pixels used.
\item Majority:   The value occuring most frequently in constituent pixels used.
\item None:   No Aggregation used.
\end{itemize}

\item file (mandatory)
        The location of the raster file. A user can navigate to the location on their file system.
        When a user is selecting an ESRI grid raster, the user should navigate to the `hdr.adf' file
        contained within the grid folder
\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item value (mandatory)


\textbf{Common connections}

\begin{itemize}
\item The output from this port only connects to the PARC input port `predictor'.
\end{itemize}

\item value\_as\_string (optional)
        This is a VisTrails port that is not used in general SAHM workflows.

\textbf{Common connections}

\begin{itemize}
\item Does not generally connect to other SAHM modules.
\end{itemize}

\end{itemize}


\section{TemplateLayer}


        The second fundamental input in an analysis is the template layer.  It is used to define the
    extent and resolution that will be used in all subsequent analyses.  The TemplateLayer is a
    raster data layer with a defined coordinate system, a known cell size, and an extent that
    defines the study area. The data type and values in this raster are not important.  All
    additional raster layers used in the analysis will be resampled and reprojected as needed to
    match the template, snapped to the template, and clipped to have an extent that matches the
    template. Users should ensure that additional covariates considered in the analysis have
    complete coverage of the template layer used.

 \subsection*{Output Ports}
\begin{itemize}
\item value (mandatory)
        This is the actual file object that is being passed to other modules in the workflow.

\textbf{Common connections}

\begin{itemize}
\item The `TemplateLayer' input port of the FieldDataAggregationAndWeight Module.
\item The `TemplateLayer' input port of the PARC Module.
\end{itemize}

\item value\_as\_string (optional)
        This is a VisTrails port that is not used in general SAHM workflows.

\textbf{Common connections}

\begin{itemize}
\item This does not commonly connect to other SAHM modules.
\end{itemize}

\end{itemize}


\section{PredictorListFile}


        The PredictorListFile module allows a user to load a .csv file containing a list of rasters
    for consideration in the modeled analysis. The .csv file should contain a header row and four
    columns containing the following information, in order, for each raster input.

        Column 1: The full file path to the input raster layer.

        Column 2: A binary value indicating whether the input layer is categorical or not.  A value
    of ``0" indicates that an input raster is non-categorical data (continuous), while a value of ``1"
    indicates that an input raster is categorical data.

        Column 3: The resampling method employed to interpolate new cell values when transforming
    the raster layer to the coordinate space or cell size of the template layer, if necessary. The
    resampling type should be specified using one of the following values: ``nearestneighbor,"
    ``bilinear," ``cubic," or ``lanczos."

        Column 4: The aggregation method to be used in the event that the raster layer must be up-
    scaled to match the template layer (e.g., generalizing a 10 m input layer to a 100 m output
    layer). Care should be taken to ensure that the aggregation method that best preserves the
    integrity of the data is used. The aggregation should be specified using one of the following
    values: ``Min," ``Mean," ``Max," ``Majority," or ``None."

        In formatting the list of predictor files, the titles assigned to each of the columns are
    unimportant as the module retrieves the information based on the order of the values in the .csv
    file (the ordering of the information and the permissible values in the file however, are
    strictly enforced). The module also anticipates a header row and will ignore the first row in
    the .csv file.



\subsection*{Input Ports}
\begin{itemize}
\item csvFileList (optional)
        This is the CSV file on the file system.  While not strictly manditory this port will almost
        always have an input.
\item predictor (optional)
        Allows a user to add individual Predictor modules to a PredictorListFile

\textbf{Common connections}

\begin{itemize}
\item The output port `value' of a Predictor module.
\end{itemize}

\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item RastersWithPARCInfoCSV (mandatory)
        This port generally connects to the input port `RastersWithPARCInfoCSV' on the PARC module.
\end{itemize}


\section{BoostedRegressionTree}


        BRT uses decision trees to partition the the parameter space into the most homogeneous
    groups in terms of the response.  BRT starts with a single decision tree, then adds a tree that
    best explains error in the first tree, and so on.  Like random forest, BRT models automatically
    model interactions and nonlinear relationships and are robust to missing observations.  Our
    implementation makes approximately 1,000 trees.  It incorporates advanced algorithms for tuning
    the model settings, simplifying the model using a cross-validation technique, and for detecting
    important interactions between covariates.  If more than 500 presence or absence records are
    found a random subset will be used for learning rate estimation and model simplification but all
    data will be used in the final model fitting step.  The cross-validation step within BRT should
    not be confused with that provided by the Model Selection Cross Validation step.  The former is
    used to optimize parameter values when defaults are not provided while the later is used to
    select models based on between model comparisons of evaluation metrics.  All discussion of
    cross-validation related to setting parameters in the BRT argument documentation refers to the
    algorithm used for parameter optimization and does not affect the cross validation split
    selected by Model Selection and Cross Validation.

        Several options are available for fitting BRTs when run using VisTrails special attention is
    required before moving away from the defaults because selection of certain parameters will
    disallow selection of others.  Optional parameters are described briefly here but a more in
    depth description can be found in Elith and Leathwich 2008.

\subsection*{Input Ports}
\begin{itemize}
\item mdsFile (mandatory)
        The the input data set consisting of locational data for each sample point, the values of
        each predictor variable at those points  This input file is almost always generated by the
        upstream steps.

\textbf{Common connections}

\begin{itemize}
\item The mdsFile can be produced by any of MDSBuilder, ModelEvaluationSplit,
ModelSelectionCrossValidation, MOdelSelectionSplit, or CorariateCorrelationAndSelection.
\end{itemize}

\item makeBinMap (optional)
        Indicate whether to discretize the continues probability map into presence absence.  See the
        ThresholdOptimizationMethod for how this is done.  If time is a concern and many models are
        to be fit and assessed maps can be produced after model selection for only the best models
        using the Select and Test the Final Model tool.  Options are available for producing
        Probability, Binary and MESS maps there as well.

\textbf{Default value} = False (Unchecked)

\textbf{Options}

\begin{itemize}
\item True (Checked)
\item False (Unchecked)
\end{itemize}

\item makeProbabilityMap (optional)
        Indicate whether a map of predicted values is to be produced for the model fit.

\textbf{Default value} = False (Unchecked)

\textbf{Options}

\begin{itemize}
\item True (Checked)
\item False (Unchecked)
\end{itemize}

\item makeMESMap (optional)
        Indicate whether to produce a multivariate environmental similarity surface (MESS) and a map
        of which factor is limiting at each point see Elith et. al. 2010 for more details.  If time
        is a concern and many models are to be fit and assessed  maps can be produced after model
        selection for only the best models using the Select and Test the Final Model tool.  Options
        are available for producing Probability, Binary and MESS maps there as well.

\textbf{Default value} = False (Unchecked)

\textbf{Options}

\begin{itemize}
\item True (Checked)
\item False (Unchecked)
\end{itemize}

\item ThresholdOptimizationMethod (optional)
        Determines how the threshold is set in order to discretize continuous predictions into
        binary. These are used for evaluation metrics calculated based on the confusion matrix as
        well as for the binary map. The options, directly from the PresenceAbsences package in R
        are:
            1:  Threshold=0.5
            2:  Sens=Spec sensitivity=specificity
            3:  MaxSens+Spec maximizes (sensitivity+specificity)/2
            4:  MaxKappa maximizes Kappa
            5:  MaxPCC maximizes PCC (percent correctly classified)
            6:  PredPrev=Obs predicted prevalence=observed prevalence
            7:  ObsPrev threshold=observed prevalence
            8:  MeanProb mean predicted probability
            9:  MinROCdist minimizes distance between ROC plot and (0,1)

        The value calculated for the train portion of the data will be applied to the test portion
        and if cross validation was specified, the value is calculated separately for each fold
        using the threshold from the training data and applying it to the test data for the hold out
        fold.

\textbf{Default value} = 2

\textbf{Options}

\begin{itemize}
\item any integer between and including 1 and 9
\end{itemize}

\item Seed (optional)
        The random number seed used by BRT.  If one desires to reproduce results from a previous BRT
        fit, one must enter the random number seed that is reported in the textual output from that
        model fit.  The seed used is always reported in the textual output.

\textbf{Default value} = Randomly Generated

\textbf{Options}

\begin{itemize}
\item Any integer between -2147483647 and 2147483647
\end{itemize}

\item TreeComplexity (optional)
        Sets the level of interactions fitted in the model.  A tree complexity of 1 fits no
        interactions, 2 will fit up to but not necessarily all two way interactions and so on.

\textbf{Default value} = If not set, tree complexity will be selected based on the number of observations and what produces the best model.

\textbf{Options}

\begin{itemize}
\item any positive integer (generally no greater than 3)
\end{itemize}

\item BagFraction (optional)
        Controls the proportion of the data that is used to fit the model at each step.  Using a bag
        fraction of 1 will give a fully deterministic model but this is generally not preferable as
        stochasticity generally improves model performance (Elith and Leathwick 2008).

\textbf{Default value} = .75

\textbf{Options}

\begin{itemize}
\item Any positive number greater than 0 and less than or equal to 1
\end{itemize}

\item NumberOfFolds (optional)
        If cross-validation is used for model simplification, this sets the number of folds used for
        cross-validation.

\textbf{Default value} = 3

\textbf{Options}

\begin{itemize}
\item A positive integer (generally between 2 and 10)
\end{itemize}

\item Alpha (optional)
        Controls when the algorithm stops in the model simplification step.  The change in deviance
        is calculated between the previous and current iteration in model simplification and if the
        average change in deviance per observation is less than the standard error of the original
        deviance multiplied by alpha then the simplification step is accepted as long as we have not
        reached the maximum number of drops allowed.

\textbf{Default value} = 1

\textbf{Options}

\begin{itemize}
\item Any positive floating point value is valid
\end{itemize}

\item PrevalenceStratify (optional)
        This specifies whether cross validation samples should be stratified to match the overall
        prevalence.  This is currently only valid for presence absence data and is only used in
        model simplification.

\textbf{Default value} = True (Checked)

\textbf{Options}

\begin{itemize}
\item True (Checked)
\item False (Unchecked)
\end{itemize}

\item ToleranceMethod (optional)
        Method used in determining when to stop model simplification.

\textbf{Default value} = ``auto"

\textbf{Options}

\begin{itemize}
\item Either ``auto" or ``fixed"
\end{itemize}

\item Tolerance (optional)
        Can be set to control the stopping rule in model simplification. If ToleranceMethod is set
        to “auto” this value will be multiplied by the mean total deviance of the null model.
        Change in deviance is compared to the tolerance to determine when to stop model
        simplification.

\textbf{Default value} = .001

\textbf{Options}

\begin{itemize}
\item Any positive floating point value is valid
\end{itemize}

\item LearningRate (optional)
        Controls the amount each tree contributes to the model.   A small learning rate restricts
        individual tree contributions to the overall model.

\textbf{Default value} = If not specified, learning rate will be determined based on the number of trees and the tree complexity.

\textbf{Options}

\begin{itemize}
\item Any positive number greater than 0 and less than 1
\end{itemize}

\item MaximumTrees (optional)
        The absolute upper limit on the total number of tress to fit.  Setting this below 5000 will
        result in an error.

\textbf{Default value} = 10,000

\textbf{Options}

\begin{itemize}
\item Any positive integer greater than 5,000
\end{itemize}

\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item modelWorkspace 
        The R workspace where all internal details regarding the fitted model are stored.  This is
        used by the Select and Test the Final Model module.

\textbf{Common connections}

\begin{itemize}
\item `modelWorkspace' port of SAHMModelOutputViewerCell for viewing the aspatial model output.
\item `modelWorkspace' port of SAHMSpatialOutpuViewerCell for viewing the spatial model output in a mini
GIS.
\end{itemize}

\item BinaryMap 
        If specified using MakeBinaryMap=True then a surface of binary predictions is produced by
        discretizing the probability map based on the selected threshold.  This map indicates
        whether one could expect each site to be occupied or unoccupied based on the model.
\item ProbabilityMap 
        If specified using MakeProbabilityMap=True then a surface of predicted values is produced
        based on the tiffs in the input .mds file and the fitted model.  These can but do not always
        indicate the probability of finding the species at a given site.  For example if model
        calibration is poor then these will not agree well with the true probabilities though
        discrimination between presence and absences might still be good.

\item ResidualsMap 
        Model residual plots show the spatial relationship between the model deviance residuals.
        Most models assume residuals will be independent thus spatial pattern in the deviance
        residuals can be indicative of a problem with the model fit and inference based on the fit.
        It can for example indicate that important predictors were not included in the model and can
        be compared with the spatial pattern of predictors that were not included in the model.
        Whether or not a significant spatial pattern exists in model residuals can at times be
        difficult to assess visually.  We hope to add correlograms of Moran`s I soon.  Unfortunately
        statistical tests based on the Moran's I statistic for residuals of binary response models
        lack statistical justification and thus cannot be used to test for a significant spatial
        pattern (Bivand 2008).   See Dormann 2007 for more discussion on evaluation of model
        residuals and spatial models that are appropriate for species distribution modeling.
        Residual plots can also be used to determine if certain observations contribute
        disproportionately to the deviance of the fitted model.  For a binary response model
        deviance residuals with absolute values greater than 2 can be indicative of a problem.
\item MessMap 
        If specified by selecting makeMESMap=True the the MESS and MoD surfaces will be produced.
        The MESS surface is the multivariate environment similarity surface and shows how well each
        point fits into the univariate ranges of the points for which the model was fit.  Negative
        values in this map indicate that the point is out of the range of the training data.  The
        MESS map takes the minimum value of a statistic calculate for each predictor and thus cannot
        diagnose hidden extrapolation as one might do using a hat matrix.  This surface is only
        calculated for variables that are selected in the model selection step within each model
        fitting algorithm so that variables that do not significantly affect the occurrence of the
        organism over the range of the training data will not be included in the MESS map even
        though these predictors might be influential to the organism outside the range in which the
        model was fit.  Random Forest never drops predictors so if one wishes to compare the MESS
        and MoD map before and after insignificant predictors were dropped, one can compare the MESS
        map of a Random Forest fit to that produced from the other model fit algorithms as long as
        they were fit using the same dataset.  See Elith et. al. 2010 for details on how the MESS
        map calculations are performed.
\item MoDMap 
        If specified by selecting makeMESMap=TRUE the the MESS and MoD surfaces will be produced.
        The MoD map is related to the MESS map and indicates which variable was furthest from the
        range over which the model was fit for each spatial location.  See Elith et. al. 2010 for
        details on how the MESS map calculations are performed.

\item modelEvalPlot 
        For binary data this will be a Receiver operating characteristic curve.  Which shows the
        relationship between sensitivity and specificity as the threshold for discretizing
        continuous predictions into presence absence is varied.  The threshold selected using the
        specified ThresholdOptimizationMethod is shown.  If a model selection test training split
        was specified the ROC curve for this will be shown in red and if a cross-validation split
        was specified ROC curves for each cross-validation fold will be overlayed with box plots
        summarizing cross-validation results.  If the model fits well both sensitivity and
        specificity should be well above the diagonal line.  If there is a strong disparity between
        the curves for the training data and either the testing split or cross validation standard
        deviation curves this can be indicative of model overfitting.  These plots and the
        evaluation metrics based on the confusion matrix describe the models ability to discriminate
        between presence and absence points.  The AUC value, or area under the ROC curve, is the
        probability that the model will rank a randomly chosen presence observation higher than a
        randomly chosen absence observation.  For count data this display will show several standard
        plots for assessment of model residuals.

\item ResponseCurves 
        Model response curves show the relationship between each predictor included in the model,
        while holding all other predictors constant at their means, and the fitted values.  MARS
        response curves are shown on a logit scale thus the response axis will not necessarily be
        bounded on the 0 to 1 interval.  BRT response curves will show response surfaces for any
        interaction terms included in the final model along with the percent relative influence.
\item Text\_Output 
        This file contains a summary of  the model fit.  The information contained here includes the
        number of presence observations (counts equal to or greater than 1 for count models), the
        number of absence points, the number of covariates that were considered by the model
        selection algorithm.  Note all of these can differ from the numbers in the original .mds due
        to incomplete records being deleted, and predictors with only one unique value being
        removed.  The random number seed is recorded if applicable which allows completely
        reproducible results as well as a summary of the model fit.  Evaluation Statistics are
        reported for the data used to fit the model as well as for the test or cross-validation
        split if applicable.  References for how to interpret most of these are ubiquitous in the
        literature but it is worth mentioning that interpretation of the calibration statistics is
        described by Pearce and Ferrier 2000 as well as Miller and Hui 1991.  Most metrics reported
        here can also be found in related graphical displays.
\end{itemize}
References:

    Bivand, R.S., Pebesma, E.J., and Gomez-Rubio, V. (2008). Applied Spatial Data Analysis with R.
        Springer New York, NY.

    Dormann, C.F., McPherson, J.M., Araujo, M.B., Bivand, R., Bolliger, J., et al. (2007). Methods to
        account for spatial autocorrelation in the analysis of species distributional data: a
        review. Ecography 30:609–28.

    Elith, J., Kearney, M., Phillips, S. (2010). The art of modeling range-shifting species. Methods
        Ecol Evol 1:330–342

    Elith, J., Leathwick, J.R. and Hastie, T. (2008). A working guide to boosted regression trees.
        Journal of Animal Ecology, 77, 802–813.

    Miller, M.E., Hui, S.L., Tierney, W.M. (1991). Validation techniques for logistic regression models.
        Statistics in Medicine 10: 1213-26

    Pearce, J., and S. Ferrier. (2000). Evaluating the predictive performance of habitat models
        developed using logistic regression. Ecological Modelling 133:225–245.

    R Development Core Team (2011). R: A language and environment for statistical computing. R
        Foundation for Statistical Computing, Vienna, Austria.

\section{RandomForest}




\subsection*{Input Ports}
\begin{itemize}
\item mdsFile (mandatory)
        The the input data set consisting of locational data for each sample point, the values of
        each predictor variable at those points  This input file is almost always generated by the
        upstream steps.

\textbf{Common connections}

\begin{itemize}
\item The mdsFile can be produced by any of MDSBuilder, ModelEvaluationSplit,
ModelSelectionCrossValidation, MOdelSelectionSplit, or CovariateCorrelationAndSelection.
\end{itemize}

\item makeBinMap (optional)
        Indicate whether to discretize the continues probability map into presence absence.  See the
        ThresholdOptimizationMethod for how this is done.  If time is a concern and many models are
        to be fit and assessed maps can be produced after model selection for only the best models
        using the Select and Test the Final Model tool.  Options are available for producing
        Probability, Binary and MESS maps there as well.

\textbf{Default value} = False (Unchecked)

\textbf{Options}

\begin{itemize}
\item True (Checked)
\item False (Unchecked)
\end{itemize}

\item makeProbabilityMap (optional)
        Indicate whether a map of predicted values is to be produced for the model fit.

\textbf{Default value} = False (Unchecked)

\textbf{Options}

\begin{itemize}
\item True (Checked)
\item False (Unchecked)
\end{itemize}

\item makeMESMap (optional)
        Indicate whether to produce a multivariate environmental similarity surface (MESS) and a map
        of which factor is limiting at each point see Elith et. al. 2010 for more details.  If time
        is a concern and many models are to be fit and assessed  maps can be produced after model
        selection for only the best models using the Select and Test the Final Model tool.  Options
        are available for producing Probability, Binary and MESS maps there as well.

\textbf{Default value} = False (Unchecked)

\textbf{Options}

\begin{itemize}
\item True (Checked)
\item False (Unchecked)
\end{itemize}

\item ThresholdOptimizationMethod (optional)
        Determines how the threshold is set in order to discretize continuous predictions into
        binary. These are used for evaluation metrics calculated based on the confusion matrix as
        well as for the binary map. The options, directly from the PresenceAbsences package in R
        are:
            1:  Threshold=0.5
            2:  Sens=Spec sensitivity=specificity
            3:  MaxSens+Spec maximizes (sensitivity+specificity)/2
            4:  MaxKappa maximizes Kappa
            5:  MaxPCC maximizes PCC (percent correctly classified)
            6:  PredPrev=Obs predicted prevalence=observed prevalence
            7:  ObsPrev threshold=observed prevalence
            8:  MeanProb mean predicted probability
            9:  MinROCdist minimizes distance between ROC plot and (0,1)

        The value calculated for the train portion of the data will be applied to the test portion
        and if cross validation was specified, the value is calculated separately for each fold
        using the threshold from the training data and applying it to the test data for the hold out
        fold.

\textbf{Default value} = 2

\textbf{Options}

\begin{itemize}
\item any integer between and including 1 and 9
\end{itemize}

\item Seed (optional)
        The random number seed used by BRT.  If one desires to reproduce results from a previous
        randomForest fit, one must enter the random number seed that is reported in the textual
        output from that model fit.  The seed used is always reported in the textual output.

\textbf{Default value} = Randomly Generated

\textbf{Options}

\begin{itemize}
\item Any integer between -2147483647 and 2147483647
\end{itemize}

\item mTry (optional)
        By default this is optimized using the tuneRF function so that OOB error is minimized.  See
        the CRAN website for more details.

\textbf{Default value} = this is optimized using the tuneRF function so that out of bag error is minimized.

\textbf{Options}

\begin{itemize}
\item A number between 1 and the total number of valid parameters used in model fitting
\end{itemize}

\item nTrees (optional)
        See the randomForest documentation on the CRAN website for details
        http://cran.r-project.org/web/packages/randomForest/index.html.

\textbf{Default value} = randomForest function default

\item nodesize (optional)
        See the randomForest documentation on the CRAN website for details
        http://cran.r-project.org/web/packages/randomForest/index.html.

\textbf{Default value} = randomForest function default

\textbf{Options}

\begin{itemize}
\item See randomForest documentation for valid input
\end{itemize}

\item replace (optional)
        See the randomForest documentation on the CRAN website for details
        http://cran.r-project.org/web/packages/randomForest/index.html.

\textbf{Default value} = randomForest function default

\textbf{Options}

\begin{itemize}
\item See randomForest documentation for valid input
\end{itemize}

\item maxnodes (optional)
        See the randomForest documentation on the CRAN website for details
        http://cran.r-project.org/web/packages/randomForest/index.html.

\textbf{Default value} = randomForest function default

\textbf{Options}

\begin{itemize}
\item See randomForest documentation for valid input
\end{itemize}

\item importance (optional)
        See the randomForest documentation on the CRAN website for details
        http://cran.r-project.org/web/packages/randomForest/index.html.

\textbf{Default value} = randomForest function default

\textbf{Options}

\begin{itemize}
\item See randomForest documentation for valid input
\end{itemize}

\item localImp (optional)
        See the randomForest documentation on the CRAN website for details
        http://cran.r-project.org/web/packages/randomForest/index.html.

\textbf{Default value} = randomForest function default

\textbf{Options}

\begin{itemize}
\item See randomForest documentation for valid input
\end{itemize}

\item proximity (optional)
        See the randomForest documentation on the CRAN website for details
        http://cran.r-project.org/web/packages/randomForest/index.html.

\textbf{Default value} = randomForest function default

\textbf{Options}

\begin{itemize}
\item See randomForest documentation for valid input
\end{itemize}

\item oobProx (optional)
        See the randomForest documentation on the CRAN website for details
        http://cran.r-project.org/web/packages/randomForest/index.html.

\textbf{Default value} = randomForest function default

\textbf{Options}

\begin{itemize}
\item See randomForest documentation for valid input
\end{itemize}

\item normVotes (optional)
        See the randomForest documentation on the CRAN website for details
        http://cran.r-project.org/web/packages/randomForest/index.html.

\textbf{Default value} = randomForest function default

\textbf{Options}

\begin{itemize}
\item See randomForest documentation for valid input
\end{itemize}

\item doTrace (optional)
        See the randomForest documentation on the CRAN website for details
        http://cran.r-project.org/web/packages/randomForest/index.html.

\textbf{Default value} = randomForest function default

\textbf{Options}

\begin{itemize}
\item See randomForest documentation for valid input
\end{itemize}

\item keepForest (optional)
        See the randomForest documentation on the CRAN website for details
        http://cran.r-project.org/web/packages/randomForest/index.html.

\textbf{Default value} = randomForest function default

\textbf{Options}

\begin{itemize}
\item See randomForest documentation for valid input
\end{itemize}

\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item modelWorkspace 
        The R workspace where all internal details regarding the fitted model are stored.  This is
        used by the Select and Test the Final Model module.

\textbf{Common connections}

\begin{itemize}
\item `modelWorkspace' port of SAHMModelOutputViewerCell for viewing the aspatial model output.
\item `modelWorkspace' port of SAHMSpatialOutpuViewerCell for viewing the spatial model output in a mini
GIS.
\end{itemize}

\item BinaryMap 
        If specified using MakeBinaryMap=True then a surface of binary predictions is produced by
        discretizing the probability map based on the selected threshold.  This map indicates
        whether one could expect each site to be occupied or unoccupied based on the model.
\item ProbabilityMap 
        If specified using MakeProbabilityMap=True then a surface of predicted values is produced
        based on the tiffs in the input .mds file and the fitted model.  These can but do not always
        indicate the probability of finding the species at a given site.  For example if model
        calibration is poor then these will not agree well with the true probabilities though
        discrimination between presence and absences might still be good.

\item ResidualsMap 
        Model residual plots show the spatial relationship between the model deviance residuals.
        Most models assume residuals will be independent thus spatial pattern in the deviance
        residuals can be indicative of a problem with the model fit and inference based on the fit.
        It can for example indicate that important predictors were not included in the model and can
        be compared with the spatial pattern of predictors that were not included in the model.
        Whether or not a significant spatial pattern exists in model residuals can at times be
        difficult to assess visually.  We hope to add correlograms of Moran`s I soon.  Unfortunately
        statistical tests based on the Moran's I statistic for residuals of binary response models
        lack statistical justification and thus cannot be used to test for a significant spatial
        pattern (Bivand 2008).   See Dormann 2007 for more discussion on evaluation of model
        residuals and spatial models that are appropriate for species distribution modeling.
        Residual plots can also be used to determine if certain observations contribute
        disproportionately to the deviance of the fitted model.  For a binary response model
        deviance residuals with absolute values greater than 2 can be indicative of a problem.
\item MessMap 
        If specified by selecting makeMESMap=True the the MESS and MoD surfaces will be produced.
        The MESS surface is the multivariate environment similarity surface and shows how well each
        point fits into the univariate ranges of the points for which the model was fit.  Negative
        values in this map indicate that the point is out of the range of the training data.  The
        MESS map takes the minimum value of a statistic calculate for each predictor and thus cannot
        diagnose hidden extrapolation as one might do using a hat matrix.  This surface is only
        calculated for variables that are selected in the model selection step within each model
        fitting algorithm so that variables that do not significantly affect the occurrence of the
        organism over the range of the training data will not be included in the MESS map even
        though these predictors might be influential to the organism outside the range in which the
        model was fit.  Random Forest never drops predictors so if one wishes to compare the MESS
        and MoD map before and after insignificant predictors were dropped, one can compare the MESS
        map of a Random Forest fit to that produced from the other model fit algorithms as long as
        they were fit using the same dataset.  See Elith et. al. 2010 for details on how the MESS
        map calculations are performed.
\item MoDMap 
        If specified by selecting makeMESMap=TRUE the the MESS and MoD surfaces will be produced.
        The MoD map is related to the MESS map and indicates which variable was furthest from the
        range over which the model was fit for each spatial location.  See Elith et. al. 2010 for
        details on how the MESS map calculations are performed.

\item modelEvalPlot 
        For binary data this will be a Receiver operating characteristic curve.  Which shows the
        relationship between sensitivity and specificity as the threshold for discretizing
        continuous predictions into presence absence is varied.  The threshold selected using the
        specified ThresholdOptimizationMethod is shown.  If a model selection test training split
        was specified the ROC curve for this will be shown in red and if a cross-validation split
        was specified ROC curves for each cross-validation fold will be overlayed with box plots
        summarizing cross-validation results.  If the model fits well both sensitivity and
        specificity should be well above the diagonal line.  If there is a strong disparity between
        the curves for the training data and either the testing split or cross validation standard
        deviation curves this can be indicative of model overfitting.  These plots and the
        evaluation metrics based on the confusion matrix describe the models ability to discriminate
        between presence and absence points.  The AUC value, or area under the ROC curve, is the
        probability that the model will rank a randomly chosen presence observation higher than a
        randomly chosen absence observation.  For count data this display will show several standard
        plots for assessment of model residuals.

\item ResponseCurves 
        Model response curves show the relationship between each predictor included in the model,
        while holding all other predictors constant at their means, and the fitted values.  MARS
        response curves are shown on a logit scale thus the response axis will not necessarily be
        bounded on the 0 to 1 interval.  BRT response curves will show response surfaces for any
        interaction terms included in the final model along with the percent relative influence.
\item Text\_Output 
        This file contains a summary of  the model fit.  The information contained here includes the
        number of presence observations (counts equal to or greater than 1 for count models), the
        number of absence points, the number of covariates that were considered by the model
        selection algorithm.  Note all of these can differ from the numbers in the original .mds due
        to incomplete records being deleted, and predictors with only one unique value being
        removed.  The random number seed is recorded if applicable which allows completely
        reproducible results as well as a summary of the model fit.  Evaluation Statistics are
        reported for the data used to fit the model as well as for the test or cross-validation
        split if applicable.  References for how to interpret most of these are ubiquitous in the
        literature but it is worth mentioning that interpretation of the calibration statistics is
        described by Pearce and Ferrier 2000 as well as Miller and Hui 1991.  Most metrics reported
        here can also be found in related graphical displays.
\end{itemize}
References:

    Bivand, R.S., Pebesma, E.J., and Gomez-Rubio, V. (2008). Applied Spatial Data Analysis with R.
        Springer New York, NY.

    Dormann, C.F., McPherson, J.M., Araujo, M.B., Bivand, R., Bolliger, J., et al. (2007). Methods to
        account for spatial autocorrelation in the analysis of species distributional data: a
        review. Ecography 30:609–28.

    Elith, J., Kearney, M., Phillips, S. (2010). The art of modeling range-shifting species. Methods
        Ecol Evol 1:330–342

    Liaw, A. and Wiener M. (2002). Classification and Regression by randomForest. R News 2(3), 18--22.

    Miller, M.E., Hui, S.L., Tierney, W.M. (1991). Validation techniques for logistic regression models.
        Statistics in Medicine 10: 1213-26

    Pearce, J., and S. Ferrier. (2000). Evaluating the predictive performance of habitat models
        developed using logistic regression. Ecological Modelling 133:225–245.

    R Development Core Team (2011). R: A language and environment for statistical computing. R
        Foundation for Statistical Computing, Vienna, Austria.

\section{MAXENT}




  

\section{MARS}


        MARS is a non-parametric technique that builds flexible models by fitting piecewise logistic
    regressions.  In effect, it is similar to GLM except that rather than fitting a straight line
    response to each predictor, piecewise functions of each predictor are fit, which allows MARS to
    better accommodate nonlinear response to predictors and also reduces the risk that outlying
    observations might have high leverage.  The model is deliberately over-fit and then pruned back.
    The original code was developed from that provided in the supporting material of Leathwick and
    Elith 2006 which contains more details on how model fitting occurs.

\subsection*{Input Ports}
\begin{itemize}
\item mdsFile (mandatory)
        The the input data set consisting of locational data for each sample point, the values of
        each predictor variable at those points  This input file is almost always generated by the
        upstream steps.

\textbf{Common connections}

\begin{itemize}
\item The mdsFile can be produced by any of MDSBuilder, ModelEvaluationSplit,
ModelSelectionCrossValidation, MOdelSelectionSplit, or CovariateCorrelationAndSelection.
\end{itemize}

\item makeBinMap (optional)
        Indicate whether to discretize the continues probability map into presence absence.  See the
        ThresholdOptimizationMethod for how this is done.  If time is a concern and many models are
        to be fit and assessed maps can be produced after model selection for only the best models
        using the Select and Test the Final Model tool.  Options are available for producing
        Probability, Binary and MESS maps there as well.

\textbf{Default value} = False (Unchecked)

\textbf{Options}

\begin{itemize}
\item True (Checked)
\item False (Unchecked)
\end{itemize}

\item makeProbabilityMap (optional)
        Indicate whether a map of predicted values is to be produced for the model fit.

\textbf{Default value} = False (Unchecked)

\textbf{Options}

\begin{itemize}
\item True (Checked)
\item False (Unchecked)
\end{itemize}

\item makeMESMap (optional)
        Indicate whether to produce a multivariate environmental similarity surface (MESS) and a map
        of which factor is limiting at each point see Elith et. al. 2010 for more details.  If time
        is a concern and many models are to be fit and assessed  maps can be produced after model
        selection for only the best models using the Select and Test the Final Model tool.  Options
        are available for producing Probability, Binary and MESS maps there as well.

\textbf{Default value} = False (Unchecked)

\textbf{Options}

\begin{itemize}
\item True (Checked)
\item False (Unchecked)
\end{itemize}

\item ThresholdOptimizationMethod (optional)
        Determines how the threshold is set in order to discretize continuous predictions into
        binary. These are used for evaluation metrics calculated based on the confusion matrix as
        well as for the binary map. The options, directly from the PresenceAbsences package in R
        are:
            1:  Threshold=0.5
            2:  Sens=Spec sensitivity=specificity
            3:  MaxSens+Spec maximizes (sensitivity+specificity)/2
            4:  MaxKappa maximizes Kappa
            5:  MaxPCC maximizes PCC (percent correctly classified)
            6:  PredPrev=Obs predicted prevalence=observed prevalence
            7:  ObsPrev threshold=observed prevalence
            8:  MeanProb mean predicted probability
            9:  MinROCdist minimizes distance between ROC plot and (0,1)

        The value calculated for the train portion of the data will be applied to the test portion
        and if cross validation was specified, the value is calculated separately for each fold
        using the threshold from the training data and applying it to the test data for the hold out
        fold.

\textbf{Default value} = 2

\textbf{Options}

\begin{itemize}
\item any integer between and including 1 and 9
\end{itemize}

\item MarsDegree (optional)
        The level of interaction allowed:
            1=no interactions (default) terms are allowed in the model
            2=1st order interactions
            3=2nd order interactions and so on.

\textbf{Default value} = 1

\textbf{Options}

\begin{itemize}
\item A positive integer generally no greater than 3 or possibly 4
\end{itemize}

\item MarsPenalty (optional)
        The cost per degree of freedom charge in fitting the mars model (from the mda library).

\textbf{Default value} = 2

\textbf{Options}

\begin{itemize}
\item A positive float
\end{itemize}

\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item modelWorkspace 
        The R workspace where all internal details regarding the fitted model are stored.  This is
        used by the Select and Test the Final Model module.

\textbf{Common connections}

\begin{itemize}
\item `modelWorkspace' port of SAHMModelOutputViewerCell for viewing the aspatial model output.
\item `modelWorkspace' port of SAHMSpatialOutpuViewerCell for viewing the spatial model output in a mini
GIS.
\end{itemize}

\item BinaryMap 
        If specified using MakeBinaryMap=True then a surface of binary predictions is produced by
        discretizing the probability map based on the selected threshold.  This map indicates
        whether one could expect each site to be occupied or unoccupied based on the model.
\item ProbabilityMap 
        If specified using MakeProbabilityMap=True then a surface of predicted values is produced
        based on the tiffs in the input .mds file and the fitted model.  These can but do not always
        indicate the probability of finding the species at a given site.  For example if model
        calibration is poor then these will not agree well with the true probabilities though
        discrimination between presence and absences might still be good.

\item ResidualsMap 
        Model residual plots show the spatial relationship between the model deviance residuals.
        Most models assume residuals will be independent thus spatial pattern in the deviance
        residuals can be indicative of a problem with the model fit and inference based on the fit.
        It can for example indicate that important predictors were not included in the model and can
        be compared with the spatial pattern of predictors that were not included in the model.
        Whether or not a significant spatial pattern exists in model residuals can at times be
        difficult to assess visually.  We hope to add correlograms of Moran`s I soon.  Unfortunately
        statistical tests based on the Moran's I statistic for residuals of binary response models
        lack statistical justification and thus cannot be used to test for a significant spatial
        pattern (Bivand 2008).   See Dormann 2007 for more discussion on evaluation of model
        residuals and spatial models that are appropriate for species distribution modeling.
        Residual plots can also be used to determine if certain observations contribute
        disproportionately to the deviance of the fitted model.  For a binary response model
        deviance residuals with absolute values greater than 2 can be indicative of a problem.
\item MessMap 
        If specified by selecting makeMESMap=True the the MESS and MoD surfaces will be produced.
        The MESS surface is the multivariate environment similarity surface and shows how well each
        point fits into the univariate ranges of the points for which the model was fit.  Negative
        values in this map indicate that the point is out of the range of the training data.  The
        MESS map takes the minimum value of a statistic calculate for each predictor and thus cannot
        diagnose hidden extrapolation as one might do using a hat matrix.  This surface is only
        calculated for variables that are selected in the model selection step within each model
        fitting algorithm so that variables that do not significantly affect the occurrence of the
        organism over the range of the training data will not be included in the MESS map even
        though these predictors might be influential to the organism outside the range in which the
        model was fit.  Random Forest never drops predictors so if one wishes to compare the MESS
        and MoD map before and after insignificant predictors were dropped, one can compare the MESS
        map of a Random Forest fit to that produced from the other model fit algorithms as long as
        they were fit using the same dataset.  See Elith et. al. 2010 for details on how the MESS
        map calculations are performed.
\item MoDMap 
        If specified by selecting makeMESMap=TRUE the the MESS and MoD surfaces will be produced.
        The MoD map is related to the MESS map and indicates which variable was furthest from the
        range over which the model was fit for each spatial location.  See Elith et. al. 2010 for
        details on how the MESS map calculations are performed.

\item modelEvalPlot 
        For binary data this will be a Receiver operating characteristic curve.  Which shows the
        relationship between sensitivity and specificity as the threshold for discretizing
        continuous predictions into presence absence is varied.  The threshold selected using the
        specified ThresholdOptimizationMethod is shown.  If a model selection test training split
        was specified the ROC curve for this will be shown in red and if a cross-validation split
        was specified ROC curves for each cross-validation fold will be overlayed with box plots
        summarizing cross-validation results.  If the model fits well both sensitivity and
        specificity should be well above the diagonal line.  If there is a strong disparity between
        the curves for the training data and either the testing split or cross validation standard
        deviation curves this can be indicative of model overfitting.  These plots and the
        evaluation metrics based on the confusion matrix describe the models ability to discriminate
        between presence and absence points.  The AUC value, or area under the ROC curve, is the
        probability that the model will rank a randomly chosen presence observation higher than a
        randomly chosen absence observation.  For count data this display will show several standard
        plots for assessment of model residuals.

\item ResponseCurves 
        Model response curves show the relationship between each predictor included in the model,
        while holding all other predictors constant at their means, and the fitted values.  MARS
        response curves are shown on a logit scale thus the response axis will not necessarily be
        bounded on the 0 to 1 interval.  BRT response curves will show response surfaces for any
        interaction terms included in the final model along with the percent relative influence.
\item Text\_Output 
        This file contains a summary of  the model fit.  The information contained here includes the
        number of presence observations (counts equal to or greater than 1 for count models), the
        number of absence points, the number of covariates that were considered by the model
        selection algorithm.  Note all of these can differ from the numbers in the original .mds due
        to incomplete records being deleted, and predictors with only one unique value being
        removed.  The random number seed is recorded if applicable which allows completely
        reproducible results as well as a summary of the model fit.  Evaluation Statistics are
        reported for the data used to fit the model as well as for the test or cross-validation
        split if applicable.  References for how to interpret most of these are ubiquitous in the
        literature but it is worth mentioning that interpretation of the calibration statistics is
        described by Pearce and Ferrier 2000 as well as Miller and Hui 1991.  Most metrics reported
        here can also be found in related graphical displays.
\end{itemize}
References:

    Bivand, R.S., Pebesma, E.J., and Gomez-Rubio, V. (2008). Applied Spatial Data Analysis with R.
        Springer New York, NY.

    Dormann, C.F., McPherson, J.M., Araujo, M.B., Bivand, R., Bolliger, J., et al. (2007). Methods to
        account for spatial autocorrelation in the analysis of species distributional data: a
        review. Ecography 30:609–28.

    Elith, J., Kearney, M., Phillips, S. (2010). The art of modeling range-shifting species. Methods
        Ecol Evol 1:330–342

    Hastie, T. and Tibshirani., R.  mda: Mixture and flexible discriminant analysis. Ported to R by
        Leisch, F., Hornik, K. and Ripley B. D.  (2011). R package version 0.4-2.

    Leathwick J.R., Elith, J., Hastie, T. (2006). Comparative performance of generalized additive models
        and multivariate adaptive regression splines for statistical modelling of species
        distributions. Ecological Modelling 199: 188-96

    Miller, M.E., Hui, S.L., Tierney, W.M. (1991). Validation techniques for logistic regression models.
        Statistics in Medicine 10: 1213-26

    Pearce, J., and S. Ferrier. (2000). Evaluating the predictive performance of habitat models
        developed using logistic regression. Ecological Modelling 133:225–245.

    R Development Core Team (2011). R: A language and environment for statistical computing. R
        Foundation for Statistical Computing, Vienna, Austria.

\section{GLM}


        This is basically linear regression adapted to binary presence-absence or count data.  We
    used a bidirectional stepwise procedure to select covariates to be used in the model.  That is,
    we began with a null model and calculated the AIC (Akaike Information Criterion) score for each
    covariate which could be added to the model.  AIC is a measure of how well the model fits the
    data with a penalty based on the number of covariates in the model.  In the first step, we add
    the covariate with the best AIC score.  In the next step we calculate AIC scores for all two-
    covariate models and again add the covariate that most improves the AIC, and so on.  At each
    step, we also look at the change in AIC from dropping each covariate currently in the model.
    The stepwise procedure ends when no additions or removals result in an improvement in AIC.



\subsection*{Input Ports}
\begin{itemize}
\item mdsFile (mandatory)
        The the input data set consisting of locational data for each sample point, the values of
        each predictor variable at those points  This input file is almost always generated by the
        upstream steps.

\textbf{Common connections}

\begin{itemize}
\item The mdsFile can be produced by any of MDSBuilder, ModelEvaluationSplit,
ModelSelectionCrossValidation, MOdelSelectionSplit, or CovariateCorrelationAndSelection.
\end{itemize}

\item makeBinMap (optional)
        Indicate whether to discretize the continues probability map into presence absence.  See the
        ThresholdOptimizationMethod for how this is done.  If time is a concern and many models are
        to be fit and assessed maps can be produced after model selection for only the best models
        using the Select and Test the Final Model tool.  Options are available for producing
        Probability, Binary and MESS maps there as well.

\textbf{Default value} = False (Unchecked)

\textbf{Options}

\begin{itemize}
\item True (Checked)
\item False (Unchecked)
\end{itemize}

\item makeProbabilityMap (optional)
        Indicate whether a map of predicted values is to be produced for the model fit.

\textbf{Default value} = False (Unchecked)

\textbf{Options}

\begin{itemize}
\item True (Checked)
\item False (Unchecked)
\end{itemize}

\item makeMESMap (optional)
        Indicate whether to produce a multivariate environmental similarity surface (MESS) and a map
        of which factor is limiting at each point see Elith et. al. 2010 for more details.  If time
        is a concern and many models are to be fit and assessed  maps can be produced after model
        selection for only the best models using the Select and Test the Final Model tool.  Options
        are available for producing Probability, Binary and MESS maps there as well.

\textbf{Default value} = False (Unchecked)

\textbf{Options}

\begin{itemize}
\item True (Checked)
\item False (Unchecked)
\end{itemize}

\item ThresholdOptimizationMethod (optional)
        Determines how the threshold is set in order to discretize continuous predictions into
        binary. These are used for evaluation metrics calculated based on the confusion matrix as
        well as for the binary map. The options, directly from the PresenceAbsences package in R
        are:
            1:  Threshold=0.5
            2:  Sens=Spec sensitivity=specificity
            3:  MaxSens+Spec maximizes (sensitivity+specificity)/2
            4:  MaxKappa maximizes Kappa
            5:  MaxPCC maximizes PCC (percent correctly classified)
            6:  PredPrev=Obs predicted prevalence=observed prevalence
            7:  ObsPrev threshold=observed prevalence
            8:  MeanProb mean predicted probability
            9:  MinROCdist minimizes distance between ROC plot and (0,1)

        The value calculated for the train portion of the data will be applied to the test portion
        and if cross validation was specified, the value is calculated separately for each fold
        using the threshold from the training data and applying it to the test data for the hold out
        fold.

\textbf{Default value} = 2

\textbf{Options}

\begin{itemize}
\item any integer between and including 1 and 9
\end{itemize}

\item SimplificationMethod (optional)
        This alters the decision rule governing how the model is pruned in the stepwise model
        selection step.

\textbf{Default value} = AIC

\textbf{Options}

\begin{itemize}
\item AIC or BIC
\end{itemize}

\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item modelWorkspace 
        The R workspace where all internal details regarding the fitted model are stored.  This is
        used by the Select and Test the Final Model module.

\textbf{Common connections}

\begin{itemize}
\item `modelWorkspace' port of SAHMModelOutputViewerCell for viewing the aspatial model output.
\item `modelWorkspace' port of SAHMSpatialOutpuViewerCell for viewing the spatial model output in a mini
GIS.
\end{itemize}

\item BinaryMap 
        If specified using MakeBinaryMap=True then a surface of binary predictions is produced by
        discretizing the probability map based on the selected threshold.  This map indicates
        whether one could expect each site to be occupied or unoccupied based on the model.
\item ProbabilityMap 
        If specified using MakeProbabilityMap=True then a surface of predicted values is produced
        based on the tiffs in the input .mds file and the fitted model.  These can but do not always
        indicate the probability of finding the species at a given site.  For example if model
        calibration is poor then these will not agree well with the true probabilities though
        discrimination between presence and absences might still be good.

\item ResidualsMap 
        Model residual plots show the spatial relationship between the model deviance residuals.
        Most models assume residuals will be independent thus spatial pattern in the deviance
        residuals can be indicative of a problem with the model fit and inference based on the fit.
        It can for example indicate that important predictors were not included in the model and can
        be compared with the spatial pattern of predictors that were not included in the model.
        Whether or not a significant spatial pattern exists in model residuals can at times be
        difficult to assess visually.  We hope to add correlograms of Moran`s I soon.  Unfortunately
        statistical tests based on the Moran's I statistic for residuals of binary response models
        lack statistical justification and thus cannot be used to test for a significant spatial
        pattern (Bivand 2008).   See Dormann 2007 for more discussion on evaluation of model
        residuals and spatial models that are appropriate for species distribution modeling.
        Residual plots can also be used to determine if certain observations contribute
        disproportionately to the deviance of the fitted model.  For a binary response model
        deviance residuals with absolute values greater than 2 can be indicative of a problem.
\item MessMap 
        If specified by selecting makeMESMap=True the the MESS and MoD surfaces will be produced.
        The MESS surface is the multivariate environment similarity surface and shows how well each
        point fits into the univariate ranges of the points for which the model was fit.  Negative
        values in this map indicate that the point is out of the range of the training data.  The
        MESS map takes the minimum value of a statistic calculate for each predictor and thus cannot
        diagnose hidden extrapolation as one might do using a hat matrix.  This surface is only
        calculated for variables that are selected in the model selection step within each model
        fitting algorithm so that variables that do not significantly affect the occurrence of the
        organism over the range of the training data will not be included in the MESS map even
        though these predictors might be influential to the organism outside the range in which the
        model was fit.  Random Forest never drops predictors so if one wishes to compare the MESS
        and MoD map before and after insignificant predictors were dropped, one can compare the MESS
        map of a Random Forest fit to that produced from the other model fit algorithms as long as
        they were fit using the same dataset.  See Elith et. al. 2010 for details on how the MESS
        map calculations are performed.
\item MoDMap 
        If specified by selecting makeMESMap=TRUE the the MESS and MoD surfaces will be produced.
        The MoD map is related to the MESS map and indicates which variable was furthest from the
        range over which the model was fit for each spatial location.  See Elith et. al. 2010 for
        details on how the MESS map calculations are performed.

\item modelEvalPlot 
        For binary data this will be a Receiver operating characteristic curve.  Which shows the
        relationship between sensitivity and specificity as the threshold for discretizing
        continuous predictions into presence absence is varied.  The threshold selected using the
        specified ThresholdOptimizationMethod is shown.  If a model selection test training split
        was specified the ROC curve for this will be shown in red and if a cross-validation split
        was specified ROC curves for each cross-validation fold will be overlayed with box plots
        summarizing cross-validation results.  If the model fits well both sensitivity and
        specificity should be well above the diagonal line.  If there is a strong disparity between
        the curves for the training data and either the testing split or cross validation standard
        deviation curves this can be indicative of model overfitting.  These plots and the
        evaluation metrics based on the confusion matrix describe the models ability to discriminate
        between presence and absence points.  The AUC value, or area under the ROC curve, is the
        probability that the model will rank a randomly chosen presence observation higher than a
        randomly chosen absence observation.  For count data this display will show several standard
        plots for assessment of model residuals.

\item ResponseCurves 
        Model response curves show the relationship between each predictor included in the model,
        while holding all other predictors constant at their means, and the fitted values.  MARS
        response curves are shown on a logit scale thus the response axis will not necessarily be
        bounded on the 0 to 1 interval.  BRT response curves will show response surfaces for any
        interaction terms included in the final model along with the percent relative influence.
\item Text\_Output 
        This file contains a summary of  the model fit.  The information contained here includes the
        number of presence observations (counts equal to or greater than 1 for count models), the
        number of absence points, the number of covariates that were considered by the model
        selection algorithm.  Note all of these can differ from the numbers in the original .mds due
        to incomplete records being deleted, and predictors with only one unique value being
        removed.  The random number seed is recorded if applicable which allows completely
        reproducible results as well as a summary of the model fit.  Evaluation Statistics are
        reported for the data used to fit the model as well as for the test or cross-validation
        split if applicable.  References for how to interpret most of these are ubiquitous in the
        literature but it is worth mentioning that interpretation of the calibration statistics is
        described by Pearce and Ferrier 2000 as well as Miller and Hui 1991.  Most metrics reported
        here can also be found in related graphical displays.
\end{itemize}
References:

    Bivand, R.S., Pebesma, E.J., and Gomez-Rubio, V. (2008). Applied Spatial Data Analysis with R.
        Springer New York, NY.

    Dormann, C.F., McPherson, J.M., Araujo, M.B., Bivand, R., Bolliger, J., et al. (2007). Methods to
        account for spatial autocorrelation in the analysis of species distributional data: a
        review. Ecography 30:609–28.

    Elith, J., Kearney, M., Phillips, S. (2010). The art of modeling range-shifting species. Methods
        Ecol Evol 1:330–342

    Miller, M.E., Hui, S.L., Tierney, W.M. (1991). Validation techniques for logistic regression models.
        Statistics in Medicine 10: 1213-26

    Pearce, J., and S. Ferrier. (2000). Evaluating the predictive performance of habitat models
        developed using logistic regression. Ecological Modelling 133:225–245.

    R Development Core Team (2011). R: A language and environment for statistical computing. R
        Foundation for Statistical Computing, Vienna, Austria.

\section{SAHMSpatialOutputViewerCell}



            SAHMModelOutputViewerCell is a VisTrails Module that
            displays the various output from a SAHM Model run in a single cell



\subsection*{Input Ports}
\begin{itemize}
\item row (optional)
        Entering a value here forces the output for this cell to appear on the row specified on the
        output spreadsheet.
        Row counts start from 1.

\textbf{Default value} = VisTrails will autoselect the next empty cell, but you do not have control over which outputs appear where.

\item column (optional)
        Entering a value here forces the output for this cell to appear on the column specified on
        the output spreadsheet.
        Row counts start from 1.
\item model\_workspace (optional)
        This is the model workspace output by any of the R models.  This widget finds all of the
        various outputs for display relative to the loction of this file.

\textbf{Common connections}

\begin{itemize}
\item Connects to the output port `modelWorkspace' on any of the R models (MARS, BRT, RandomForest, or
GLM)
\end{itemize}

\item max\_cells\_dimension (optional)
        For very large GIS datasets the Matplotlib based viewer used to display GIS data runs into
        memory issues.  To get around this limitation we downsample the full resolution and only
        display the resampled image.  This number represents the number of resampled pixels that
        will be displayed in the largest dimension.  For example if your template layer is 20,000 by
        10,000 pixels and the max\_cells\_dimension is set at 5000 your output will be displayed at
        5,000 by 2,500 resolution.  Zooming in the map does cause the resampled cells to be
        displayed at their original resolution as is the case in most GIS programs.  If the
        resampled resolution is not sufficient for your needs you will need to display the final
        outputs in an external GIS program.

\textbf{Default value} = 5000

\end{itemize}

 

\section{SAHMModelOutputViewerCell}


        The SAHM Spatial Output Viewer Cell provides a convenient means for viewing the numerous
    spatial outputs produced by individual model runs as well as the input presence and absence
    points and background points if applicable.  The spatial viewer displays the outputs in an
    interactive Matplotlib chart which functions much like a full GIS.

        Attached to each cell is a toolbar that allows changing of the displayed layer and the
    overlaid points

\subsection*{Input Ports}
\begin{itemize}
\item row (optional)
        Entering a value here forces the output for this cell to appear on the row specified on the
        output spreadsheet.
        Row counts start from 1.

\textbf{Default value} = VisTrails will autoselect the next empty cell, but you do not have control over which outputs appear where.

\item column (optional)
        Entering a value here forces the output for this cell to appear on the column specified on
        the output spreadsheet.
        Row counts start from 1.
\item model\_workspace (optional)
        This is the model workspace output by any of the R models.  This widget finds all of the
        various outputs for display relative to the loction of this file.

\textbf{Common connections}

\begin{itemize}
\item Connects to the output port `modelWorkspace' on any of the R models (MARS, BRT, RandomForest, or
GLM)
\end{itemize}

\item InitialModelOutputDisplay (optional)
        The display tab to show initially.

\textbf{Default value} = AUC Curves

\textbf{Options}

\begin{itemize}
\item Text
\item Response Curves
\item AUC Curves
\item Calibration
\item Confusion
\item Residuals
\end{itemize}

\end{itemize}

 

\section{FieldDataQuery}


        Often raw field data come to us in a format that contains more information than we need to
    include in any single model.  This can take the form of additional columns that contain
    extraneous information, additional columns that contain occurrence data for additional species,
    or rows that from a time period, collection method, or species that we are not interested in
    modeling.  The Field Data Query module contains functionality to subset and reformat this output
    into the format used by the SAHM package.
            Columns can be specified with either a positional argument (1, 2, 3, etc) if you want to
    select the first, second, third etc column.  Note these numbers start from 1.  Alternatively you
    can select a column based on name by entering the text of the column name found in the header.

            When selecting rows there are two types of queries that can be specified:
               Simple - Select a Query\_Column and enter a value in the Query port.  If the value for
    a row in the selected column equals the value entered in the Query that row will be kept.  For
    example you might have a `year' column and you would want to select all 2009 entries.  NOTE:  DO
    NOT ENCLOSE THE QUERY TEXT IN QUOTES IF YOU ARE TRYING TO MATCH A STRING!

               Complex - Optionally you can construct complex queries using Python syntax.  To do
    this enclose the column name in square brackets as part of a line of Python code.  Since the
    columns used are specified in the query string there is no need to use the Query\_column port and
    it will be ignored.  For example to select years greater than 2005 you would use:  [year] > 2005
    To include string equality make sure you enclose the entire bracketed field name in quotes as
    well.  For example ``[Observer]" == ``Colin"  Complex queries involving multiple columns are
    possible as well, for example ``[Observer]" != ``Colin" and [year] > 2005.

\subsection*{Input Ports}
\begin{itemize}
\item fieldData\_file (mandatory)
        The file containing Field data.  The acceptable formats vary but it must have a column with
        X, Y, and response values.  Additional columns are permissible.

\textbf{Common connections}

\begin{itemize}
\item This port can be connected to a FieldData module or the FieldData file can be specified directly in
the module information pane.
\end{itemize}

\item x\_column (optional)
        The column that contains the `X' coordinates.  These values must be in the same coordinates,
        projection, and units as those defined in the template layer.

        Columns can be specified with either a positional argument (1, 2, 3, etc) if you want to
        select the first, second, third etc column.  Note these numbers start from 1.  Alternatively
        you can select a column based on name by entering the text of the column name found in the
        header.

\textbf{Default value} = 1, which is to say the first column in the input field data file.


\textbf{Common connections}

\begin{itemize}
\item NA
\end{itemize}

\item y\_column (optional)
        The column that contains the `Y' coordinates.  These values must be in the same coordinates,
        projection, and units as those defined in the template layer.

        Columns can be specified with either a positional argument (1, 2, 3, etc) if you want to
        select the first, second, third etc column.  Note these numbers start from 1.  Alternatively
        you can select a column based on name by entering the text of the column name found in the
        header.

\textbf{Default value} = 2, which is to say the second column in the input field data file.

\item Response\_column (optional)
        The column that contains the response of interest.

        Columns can be specified with either a positional argument (1, 2, 3, etc) if you want to
        select the first, second, third etc column.  Note these numbers start from 1.  Alternatively
        you can select a column based on name by entering the text of the column name found in the
        header.

\textbf{Default value} = 3, which is to say the third column in the input field data file.

\item ResponseType (optional)
        The type of response recorded in the response column

\textbf{Default value} = `Presence(Absence)'

\textbf{Options}

\begin{itemize}
\item `Presence(Absence)' = 1 for Presense, optionally also 0 for Absence and -9999 for background points.
\item `Count' = 0, 1, 2, 3 etc. observed count data.  Optionally also -9999 for background points
\end{itemize}

\item Response\_Presence\_value (optional)
        The value in the response column that will be taken to indicate a presense.

\textbf{Default value} = By default any value in the list `1', `True', `T', `Present', `Presence' will be assigned a value of 1 (presence) in the output. Note: not case sensitive.

\textbf{Options}

\begin{itemize}
\item And number or string can be entered, quotes are not required.
\end{itemize}

\item Response\_Absence\_value (optional)
        The value in the response column that will be taken to indicate an absence.

\textbf{Default value} = By default any value in the list `0', `False', `F', `Absent', `Absence' will be assigned a value of 0 (absence) in the output. Note: not case sensitive.

\textbf{Options}

\begin{itemize}
\item And number or string can be entered, quotes are not required.
\end{itemize}

\item Query\_column (optional)
        The column which contains values you would like to use to with the simple equality query
        option.  The values in this column will be checked against the value entered in the query
        port.
\item Query (optional)
        If using the simple equality query functionality simply enter the value you would like to
        filter on here.  NOTE:  DO NOT ENCLOSE THE QUERY TEXT IN QUOTES IF YOU ARE TRYING TO MATCH A
        STRING!  Also do not include any additional spaces.

        If using the complex Python syntax query a valid Python equality statement with the values
        from each individual row indicated with square bracketed field (header row) names.
\textbf{Options}

\begin{itemize}
\item Simple - Select a Query\_Column and enter a value in the Query port.  If the value for a row in the
selected column equals the value entered in the Query that row will be kept.  For example you might
have a `year' column selected in the Query\_column port and enter 2009 here to to select all 2009
entries.  NOTE:  DO NOT ENCLOSE THE QUERY TEXT IN QUOTES IF YOU ARE TRYING TO MATCH A STRING!
\item Complex - Optionally you can construct complex queries using Python syntax.  To do this enclose the
column name in square brackets as part of a line of Python code.  Since the columns used are
specified in the query string there is no need to use the Query\_column port and it will be ignored.
For example to select years greater than 2005 you would use:  [year] > 2005  To include string
equality make sure you enclose the entire bracketed field name in quotes as well.  For example
``[Observer]" == ``Colin"  Complex queries involving multiple columns are possible as well, for
example ``[Observer]" != ``Colin" and [year] > 2005.
\end{itemize}

\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item fieldData (mandatory)
        The file containing field data.
        This output file will be in: X, Y, ResponseBinary/ResponseCount format.

\textbf{Common connections}

\begin{itemize}
\item FieldDataAggregateAndWeight FieldData - For collapsing or weighting points relative to the template
pixels.
\item MDS\_builder - fieldData - for modeling without using the FieldDataAggregateAndWeight functionality.
\end{itemize}

\end{itemize}


\section{FieldDataAggregateAndWeight}


        In many instances data collected in the field can be at a different projection and spatial
    resolution than we are modeling at.  For example we might have observations collected every five
    meters along a 200 m. transect when we are modeling with covariates with 1000 m. cells.  When
    running species distribution models (SDMs) such as those contained in SAHM, spatial issues need
    to be addressed in order to avoid introduction of pseudo-replication.   For instance,
    considering multiple field data observations which are all spatially located in the same modeled
    pixel will generate replicate values or redundant information.  When running a model, this
    redundancy causes pseudo-replication and can negatively influence model development.  The
    FieldDataAggregateAndWeight tool helps aggregate field data locations so only one field data
    observation is represented per pixel or multiple points are downweighted proportionately.
    Additonally the FieldDataAggregationAndWeight module allows the user to change the datum /
    projection system of the FieldData x, y locations to match that used in the template.

        Currently only GLM, MARS, and Boosted Regression Trees accept weights.  Any Weights column
    will be ignored by Random Forest.


\subsection*{Input Ports}
\begin{itemize}
\item templateLayer (mandatory)
        Raster file used to determine cell size and extent.
        Note - The projection and coordinate system used in the template file must match that given
        in the FieldData`s X and Y columns.

\textbf{Common connections}

\begin{itemize}
\item This input port generally will connnect to the 'value` port of a TemplateLayer Module.
\end{itemize}

\item fieldData (mandatory)
        The file containing field data.  Must be in X, Y, ResponseBinary/ResponseCount format

\textbf{Common connections}

\begin{itemize}
\item The'value` port of a FieldData module
\item The 'fieldData` port of the FieldDataQuery module
\end{itemize}

\item PointAggregationOrWeightMethod (mandatory)
        The method used to either weight or aggregate field data points.

\textbf{Default value} = Collapse In Pixel

\textbf{Options}

\begin{itemize}
\item Collapse In Pixel : All field data points falling within a single pixel will be collapse into a
single point at the center of that pixel.  If using Presense(Absense) data the point will be given a
value of 1 if any are presense, otherwise 0 if any are absense, otherwise -9999 if all are
background.  If using count data the point value will be the average of all points in a pixel.
\item Weight Per Pixel : All field data points are retained but a weight column is added and points in
pixels with multiple points are given a weight of 1 over the number of points in that pixel.  For
example all the points in a pixel with 4 points would be given a weight of 1/4.
\end{itemize}

\item FD\_EPSG\_projection (optional)
        This optional parameter is a means of specifying the datum and projection information that
        the field data X and Y locations are in.  If this parameter is supplied the point locations
        will be transformed to the datum projection of the template layer.  Otherwise it will be
        assumed that the points and template are in the same projection and datum.  The value to
        enter in this port must be a valid EPSG code, see below.

        EPSG codes are numbers representing all commonly used geographic and projected coordinate
        systems.  These are generally 4 to 6 digit numbers.  For example 4326 represents geographic
        WGS84 data, 4260 represents geographic NAD83, 26912 represents NAD83 / UTM zone 12N, etc.
        See the options below for a list of ways to lookup EPSG codes.

\textbf{Default value} = None, This assumes that the points are in the same coordinate system as your template layer.

\textbf{Options}

\begin{itemize}
\item Within your GDAL installation directory there is a gdal-data directory with two csv files which list
all of the supported EPSG numbers and the name and information of the coordinate system.  Use
gcs.csv for geographic systems and pcs.csv for projected systems.
\item EPSG codes can also be looked up at: http://spatialreference.org/ref/
\item If you have a .prj file or well known text (WKT) for your coordinate system you can find the EPSG
using: http://prj2epsg.org/search
\item If you have ESRI ArcGIS and a layer in the coordinate system you can find the EPSG code in the items
metadata.  Select the item in ArcCatalog, open the description tag, scroll down to the Spatial
Reference section, the EPSG will be the number in the 'WELL-KNOWN-IDENDIFIER` tag.
\end{itemize}

\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item fieldData (mandatory)
        This is a CSV file in a X,Y,Response,(Weight) format.

\textbf{Common connections}

\begin{itemize}
\item The input port 'fieldData` of the MDSBuilder module.
\end{itemize}

\end{itemize}


\section{PARC}


        The Projection, Aggregation, Resampling, and Clipping (PARC) module is a powerful utility
    that automates the preparation steps required for using raster layers in most geospatial
    modeling packages. In order to successfully consider multiple environmental predictors in raster
    format, each layer must have coincident cells (pixels) of the same size, have the same
    coordinate system (and projection, if applicable), and the same geographic extent. The PARC
    module ensures that all of these conditions are met for the input layers by transforming and or
    reprojecting each raster to match the coordinate system of the template layer. This process
    usually involves aggregation (necessary when an input raster layer must be up-scaled to match
    the template layer-- e.g., generalizing a 10 m input layer to a 100 m output layer), and or
    resampling (necessary for interpolating new cell values when transforming the raster layer to
    the coordinate space or cell size of the template layer). Lastly, each raster predictor layer is
    clipped to match the extent of the template layer.

        The settings used during these processing steps follow a particular set of decision rules
    designed to preserve the integrity of data as much as possible. However, it is important for a
    user to understand how these processing steps may modify the data inputs. For additional
    information about the PARC module, please see the extended help and documentation for the SAHM
    package.

\subsection*{Input Ports}
\begin{itemize}
\item predictor (optional)
        A single raster with resampling, aggregation, and categorical options.

\textbf{Common connections}

\begin{itemize}
\item value' port of a Predictor module
Note - Multiple single Predictor modules can be connected to this single input port.
\end{itemize}

\item PredictorList (optional)
        This is an in memory data construct that contains a list of predictors each with resampling,
        aggregation, and categorical options.

\textbf{Common connections}

\begin{itemize}
\item value` port of any of the 'Individual Predictors selector` modules
Note - Multiple single Predictors selectors modules can be connected to this single input port.
\end{itemize}

\item RastersWithPARCInfoCSV (optional)
        This is a CSV containing a list of files to include in the PARC operation.

        The format of this list conforms to the 'PredictorListFile` specs:
            Column 1: The full file path to the input raster layer including the drive.
            Column 2: A binary value indicating whether the input layer is categorical or not. A
        value of ``0" indicates that an input raster is non-categorical data (continuous), while a
        value of ``1" indicates that an input raster is categorical data.
            Column 3: The resampling method employed to interpolate new cell values when
        transforming the raster layer to the coordinate space or cell size of the template layer, if
        necessary. The resampling type should be specified using one of the following values:
        ``nearestneighbor," ``bilinear," ``cubic," or ``lanczos."
            Column 4: The aggregation method to be used in the event that the raster layer must be
        up-scaled to match the template layer (e.g., generalizing a 10 m input layer to a 100 m
        output layer). Care should be taken to ensure that the aggregation method that best
        preserves the integrity of the data is used. The aggregation should be specified using one
        of the following values: ``Min," ``Mean," ``Max," ``Majority," or ``None."

        In formatting the list of predictor files, the titles assigned to each of the columns are
        unimportant as the module retrieves the information based on the order of the values in the
        .csv file (the ordering of the information and the permissible values in the file however,
        are strictly enforced). The module also anticipates a header row and will ignore the first
        row in the .csv file.

\textbf{Common connections}

\begin{itemize}
\item 'value` port of PredictorListFile module
\end{itemize}

\item templateLayer (mandatory)
        The templat layer raster file used to define the Extent, Cell size, Projection, raster snap,
        and coordinate system of the outputs.

\textbf{Common connections}

\begin{itemize}
\item 'value` port of TemplateLayer module
\end{itemize}

\item ignoreNonOverlap (optional)
        Option of using the intersection of all covariates and template or enforcing the template
        extent.
\textbf{Options}

\begin{itemize}
\item True (checked) = Use intersection of all covariates extents.  Area of template extent will be reduce
such all covariate layers extents can be completely covered by the new extent.
\item False (Unchecked) = The template extent will be used for all outputs and an error will be raised if
any of the covariates are not completely covered by the template.
\end{itemize}

\item multipleCores (optional)
        Option of running processing on multiple threads/cores.
\textbf{Options}

\begin{itemize}
\item True (checked) = Individual layers will be run consectively on separate threads.
\item False (Unchecked) = All processing will occur on the same thread as the main program.
\end{itemize}

\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item RastersWithPARCInfoCSV (mandatory)
        The VisTrails output from the PARC module is a interim CSV file that contains information
        about each of the files processed.  This is used by the MDS builder to determine which files
        to extract values from and which layers are categorical.

\textbf{Common connections}

\begin{itemize}
\item Input port 'RastersWithPARCInfoCSV` of MDSBuilder module
\end{itemize}

\end{itemize}
References:

    NA

\section{RasterFormatConverter}


        The RasterFormatConverter module allows a user to easily convert between raster file types
    for a group of rasters. This group can be specified as either all the rasters in a single
    directory or the rasters specified in a single MDS file (see below).  All outputs will be sent
    to a folder named ``ConvertedRasters" (followed by an underscore and a number corresponding to
    the run sequence of the module) within the user's current VisTrail`s session folder.  Typically
    this module will be used within a workflow to convert the geotiff format used by the rest of the
    modules to the ascii format needed by Maxent.  But the following file formats are accepted for
    both inputs and outputs: Arc/Info ASCII Grid, ESRI BIL, ERDAS Imagine, and JPEG and others.  See
    the compiled by default options at http://www.gdal.org/formats\_list.html for a complete list of
    the accepted file types.

\subsection*{Input Ports}
\begin{itemize}
\item inputMDS (optional)
        Any merged dataset (MDS) format csv can be used as input to this module.  All of the rasters
        pointed to in the third line of the file will be converted to the output format.

\textbf{Common connections}

\begin{itemize}
\item This can be connected to the output 'mdsFile` port on any of the following modules: MDSBuilder,
ModelEvaluationSplit, ModelSelectionSplit, ModelCrossValidationSplit, or
CovariateCorrelationAndSelection.
\end{itemize}

\item inputDir (optional)
        An directory can be used to specify which files to process.  All of the rasters (of any
        format) will be converted to the output format specified.

\textbf{Common connections}

\begin{itemize}
\item This does not generally connect to any other SAHM modules.
\end{itemize}

\item format (optional)
        The format to convert all the input grids into.  For Maxent this will be ASCII.

\textbf{Default value} = Geotif

\textbf{Options}

\begin{itemize}
\item Geotif
\item Arc/Info Grid
\item ASCII Grid
\item ESRI Bil
\item ERDAS Imagine
\item JPEG
\item BMP
\item Additional uncommon file types are supported by GDAL.  For the complete list see the 'compiled by
default` options at http://www.gdal.org/formats\_list.html for a complete list of the accepted file
types.
\end{itemize}

\item multipleCores (optional)
        Option of running processing on multiple threads/cores.
\textbf{Options}

\begin{itemize}
\item True (checked) = Individual layers will be run consectively on separate threads.
\item False (Unchecked) = All processing will occur on the same thread as the main program.
\end{itemize}

\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item outputDir (optional)
        The directory where all output files will be saved to.

\textbf{Default value} = This directory name is created by the module and it will be located in the session folder.


\textbf{Common connections}

\begin{itemize}
\item This port will connect to the maxent input port 'projectionlayers`.
\end{itemize}

\end{itemize}


\section{MDSBuilder}


        The Merged Data Set (MDS) Builder module is a utility that extracts the values of each
    predictor layer to the point locations included in the field data set. The module produces a
    .csv file that contains the x and y locations of the sample points and a column indicating
    whether each point represents a presence recording, an absence recording, a presence count, or a
    background point.  Following these first three columns, each environmental predictor layer is
    appended as a column with row entries representing the value present in the raster layer at each
    field sample point.  There are a total of three header rows in the output .csv of the
    MDSBuilder. The first row contains the columns ``x," ``y," ``ResponseBinary" or ``ResponseCount,"
    and the names of each of the raster predictor files that were passed to the MDS Builder. The
    second row contains a binary value indicating whether the column should be included when the
    model is finally applied; these values are later modified during the Covariate Correlation and
    Selection process that takes place downstream in the workflow. The final header row contains the
    full path on the file system to each of the raster predictor files.

        The output from this module is in the format expected by most of the pre-modeling data
    manipulation modules, all of the model modules, as well as the RasterFormatConverter module.  As
    such it can reasonabily be connected to numerous other modules depending on the type of modeling
    being conducted.  A typical workflow would linearly connect MDSBuilder -> ModleEvaluationSplit
    -> ModelSelectionSplit or ModelSelectionCrossValidation -> CovariateCorrelationAndSelection ->
    any or all of the models (BoostedRegressionTree, GLM, MARS, RandomForest, Maxent).  If using
    Maxent the output from CovariateCorrelationAndSelection would also go into the
    RasterFormatConverter module which woulc connect to the projectionlayers of the maxent module.

\subsection*{Input Ports}
\begin{itemize}
\item RastersWithPARCInfoCSV (mandatory)
        This is a csv file which contains information about all of the predictors used. The user
        will not generally need to create or edit this file as it is an output of the PARC module.

        The folowing columns are in a RastersWithPARCInfoCSV:
          PARCOutputFile - The raster file produced by PARC
          Categorical - 0=not categorical data, 1=categorical data
          Resampling - One of NearestNeighbor, bilinear, cubic, cubicspline, or lanczos
          Aggregation - One of min, max, mean, or majority
          OriginalFile - The location and name of the input file used by PARC

\textbf{Common connections}

\begin{itemize}
\item This port will generally connect with the ouput from the PARC module.
\end{itemize}

\item fieldData (mandatory)
        The field data input corresponds to a .csv file containing presence/absence points or count
        data recorded across a landscape for the phenomenon being modeled (e.g., plant sightings,
        evidence of animal presence, etc.).

\textbf{Common connections}

\begin{itemize}
\item The output port of the FieldData Module
\item The output port of the FieldDataQuery Module
\item The output port of the FieldDataAggregationAndWeight Module
\end{itemize}

\item backgroundPointCount (optional)
        This is an optional value that specifies how many randomly placed packground points to add
        to the output.  These points will be randomly placed at pixel centroids within the template
        extent with no more than one point assigned to any one pixel. In typical SAHM workflows
        these points are only used by the Maxent modeling package.  These points will be added to
        the output .csv file with a value of ``-9999" denoting them as background points.

\textbf{Default value} = 0, which is to say that no background point are added to the output.

\item backgroundProbSurf (optional)
        Background Probability Surface: This is an optional parameter that applies only to workflows
        that employ the Maxent modeling package. In some analyses, it may be appropriate to
        spatially limit background points to a particular subset of the study area (e.g., islands
        within a study area polygon, particular regions within a study area polygon, or a region
        determined by the known bias present in the field data).  Specifying a background
        probability surface raster allows a user to control where random points will be scattered
        within the extent of the study area. The raster layer specified by a user should have the
        same projection and extent as the template layer and contain values ranging from 0 to 100.
        These values represent the probability that a randomly generated point will be retained
        should it fall within a particular cell.  That is, randomly generated points will not be
        generated in any part of the probability grid with a value of ``0" while all points falling
        in an area with a value of ``100" will be retained. A point falling in an area with a value
        of ``50" will be kept as a background point 50\% of the time.

\textbf{Default value} = 0 (No background points are added to the output.)

\textbf{Options}

\begin{itemize}
\item Any positive integer.
\end{itemize}

\item Seed (optional)
        The seed is used to be able recreate a specific output.  The seed used in each run will be
        noted on the console output and saved in the output log in the session folder.

\textbf{Default value} = If no seed is specified a random seed between -1*$((2^32)/2)$ and $((2^32)/2)$ will be generated and used.

\textbf{Options}

\begin{itemize}
\item Any integer between -1*$((2^32)/2)$ and $((2^32)/2)$
\end{itemize}

\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item mdsFile (optional)
        This is the CSV flat file containing the location data and values extracted from each of the
        covariates.

\textbf{Common connections}

\begin{itemize}
\item The input port 'InputMDS` in the ModelEvaluationSplit module.
\item The input port 'InputMDS` in the ModelSelectionSplit module.
\item The input port 'InputMDS` in the ModelSelectionCrossValidation module.
\item The input port 'InputMDS` in the CovariateCorrelationAndSelection module.
\item The input port 'InputMDS` in the RasterFormatConverter module.
\item The input port 'InputMDS` in the Maxent module.
\item The input port 'InputMDS` in the BoostedRegressionTree module.
\item The input port 'InputMDS` in the GLM module.
\item The input port 'InputMDS` in the MARS module.
\item The input port 'InputMDS` in the RandomForest module.
\end{itemize}

\end{itemize}


\section{ModelSelectionSplit}


        The ModelSelectionSplit reserves a portion of the data from the model fitting process but
    reports the evaluation metrics on all models not just the those selected as the final models to
    be reported in the analysis (in contrast to the ModelEvaluationSplit).  This module should be
    placed directly the CovariateCorrelationAndSelection.  If both a ModelEvaluationSplit and a
    ModelSelectionSplit are specified then the training portion of the ModelEvalutationSplit will be
    further partitioned by the ModelSelectionSplit thus the ModelEvalutationSplit should come first
    in the workflow.  Both of these algorithms stratify the splits by the response.  That is, the
    ratio of presence to absence points should be nearly equal in the testing and training split.
    If a ModelSelectionSplit is included evaluation metrics applied to the reserved data will be
    reported in the textual output, model evaluation plots including AUC plots as well as the across
    model plots and the csv.  Both of these modules ignore background points and treat all
    observations with values greater than 0 as presence for the purpose of stratification by
    response.

        It is not valid to select models based on their performance on the reserved portion of the
    data and then report these metrics only for the top performing models claiming that we would
    expect similar performance on an independent dataset see Hastie 2009 for this discussion.  If
    one desires metrics for how the models might be expected to perform on an independent dataset
    then the ModelEvaluationSplit must be used.



\subsection*{Input Ports}
\begin{itemize}
\item inputMDS (mandatory)
        This is the input data set consisting of location data for each sample point, the values of
        each predictor variable at those points, and if established, a field denoting the weight
        that will be assigned to each point in modeling. This input is usually provided by the
        upstream steps that precede the Test Training Split module. Any value entered here (e.g.,
        specifying another existing MDS on the file system) will override the input specified by a
        model connection in the visual display.

\textbf{Common connections}

\begin{itemize}
\item This port can connect to the output port on MDSBuilder, ModelEvaluationSplit
\item While it could technically also connect to the output from ModelSelectionCrossValidation,
CovariateCorrelationAndSelection, or even another ModelSelectionSplit this would not make sense for
SAHM workflows.  The results of connecting to one of these modules has not been tested and could
cause errors or subtle but significant problems with subsequent modeling.
\end{itemize}

\item trainingProportion (optional)
        This is the proportion of the sample points that will be used to train the model, relative
        to the total number of points. Entered values should be greater than 0 but less than 1. For
        example, a value of '0.9` will result in 90\% of the sample points being used to train the
        model, with 10\% of the sample being held out to test the model's performance. Choosing an
        appropriate training proportion can depend on various factors, such as the total number of
        sample points available.  Selecting an appropriate value for the training proportion is a
        complex issue that depends on many factors including the total number of observations, the
        complexity of the models that will be fit, and the signal to noise ratio in the data (Hastie
        et. al. 2009).
\item RatioPresAbs (optional)
        This optional field is populated with a number corresponding to the desired ratio of
        presence to  absence points to be used in the analysis. If not populated then all occurrence
        records (not background points) will be portioned into either the test or training split
        with no reduction in the total number of points.  If populated, this entry should be a
        number greater than zero. (A value of `1' will result in an equal number of both presence
        and absence points being used, a value of `2' indicates that twice as many presence points
        will be used, a value of ‘0.5` indicates that twice as many absence points will be used,
        etc.). All field data points with a value equal to or greater than 1 are interpreted as
        presence points. Although the original field data is unmodified, this option will reduce the
        sample size as the merged dataset containing sample points will have points deleted from it
        to achieve the specified ratio as such it should be used with caution. A warning will be
        generated if more than 50\% of either the presence or absence points will be deleted based on
        the ratio specified by the user. Background points are ignored by this module (they are read
        in and written out, but not assigned to either the test or training split).
\item Seed (optional)
        The random number seed used by split the data.  If one desires to reproduce results from a
        previous split fit, one must enter the random number seed that is reported in the console
        output from that split step.

\textbf{Default value} = Randomly selected

\textbf{Options}

\begin{itemize}
\item Any integer between -2147483647 and 2147483647
\end{itemize}

\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item outputMDS 
        This is an MDS file to be further used in the downstream workflow.  It is nearly identical
        to the input MDS file but with an added column under the header ``Split" for each non-
        background observation will be labeled either ``test", ``train", or ``NA" indicating whether
        the observation will be used for producing evaluation metrics, training the model, or
        excluded from the analysis respectively. ``NA" generally indicates that the observation was
        reserved for evaluating the final models but can also occur if a desired ratio of presence
        to absence was set using the RatioPresAbs.

\textbf{Common connections}

\begin{itemize}
\item This port can connect to either the CovariateCorrelationAndSelection or directly to any of the R
model modules (BoostedRegressionTree, GLM, MARS, or RandomForest).
\item While it could technically also connect to the input port of ModelEvaluationSplit,
ModelSelectionCrossValidation or even another ModelSelectionSplit this would not make sense for SAHM
workflows.  The results of connecting to one of these modules has not been tested and could cause
errors or subtle but significant problems with subsequent modeling.
\end{itemize}

\end{itemize}
References:

    Hastie T, Tibshirani R, Friedman JH. 2009. The Elements of Statistical Learning: Data Mining,
        Inference, and Prediction. New York: Springer-Verlag. 744 pp. 2nd ed.

\section{ModelEvaluationSplit}


        The ModelEvaluationSplit module provides the opportunity to reserve a specified portion of
    the data for producing and reporting evaluation metrics on an independent test set following
    model exploration and selection.  The ModelEvaluationSplit must be applied before the
    CovariateCorrelationAndSelection module. The nearly identical ModelSelectionSplit reserves a
    portion of the data from the model fitting process but reports the evaluation metrics on all
    models not just the those selected as the final models to be reported in the analysis.  This
    module can be placed either directly before or directly after the
    CovariateCorrelationAndSelection.  If both a ModelEvaluationSplit and a ModelSelectionSplit are
    specified then the training portion of the ModelEvalutationSplit will be further partitioned by
    the ModelSelectionSplit thus the ModelEvalutationSplit should come first in the workflow.  Both
    of these algorithms stratify the splits by the response. That is, the ratio of presence to
    absence points should be nearly equal in the testing and training split.  If a
    ModelSelectionSplit is included evaluation metrics applied to the reserved data will be reported
    in the textual output, model evaluation plots including AUC plots as well as the across model
    plots and the csv.  Both of these modules ignore background points and treat all observations
    with values greater than 0 as presence for the purpose of stratification by response.




\subsection*{Input Ports}
\begin{itemize}
\item inputMDS (mandatory)
        This is the input data set consisting of location data for each sample point, the values of
        each predictor variable at those points, and if established, a field denoting the weight
        that will be assigned to each point in modeling. This input is usually provided by the
        upstream steps that precede the Test Training Split module. Any value entered here (e.g.,
        specifying another existing MDS on the file system) will override the input specified by a
        model connection in the visual display.

\textbf{Common connections}

\begin{itemize}
\item This port can connect to the output port on MDSBuilder.
\item While it could technically also connect to the output from ModelSelectionCrossValidation,
CovariateCorrelationAndSelection, ModelSelectionSplit, or another MOdelEvaluationSplit this would
not make sense for SAHM workflows.  The results of connecting to one of these modules has not been
tested and could cause errors or subtle but significant problems with subsequent modeling.
\end{itemize}

\item trainingProportion (optional)
        This is the proportion of the sample points that will be used to train the model, relative
        to the total number of points. Entered values should be greater than 0 but less than 1. For
        example, a value of '0.9` will result in 90\% of the sample points being used to train the
        model, with 10\% of the sample being held out to test the model's performance. Choosing an
        appropriate training proportion can depend on various factors, such as the total number of
        sample points available.  Selecting an appropriate value for the training proportion is a
        complex issue that depends on many factors including the total number of observations, the
        complexity of the models that will be fit, and the signal to noise ratio in the data (Hastie
        et. al. 2009).
\item RatioPresAbs (optional)
        This optional field is populated with a number corresponding to the desired ratio of
        presence to  absence points to be used in the analysis. If not populated then all occurrence
        records (not background points) will be portioned into either the test or training split
        with no reduction in the total number of points.  If populated, this entry should be a
        number greater than zero. (A value of `1' will result in an equal number of both presence
        and absence points being used, a value of `2' indicates that twice as many presence points
        will be used, a value of ‘0.5` indicates that twice as many absence points will be used,
        etc.). All field data points with a value equal to or greater than 1 are interpreted as
        presence points. Although the original field data is unmodified, this option will reduce the
        sample size as the merged dataset containing sample points will have points deleted from it
        to achieve the specified ratio as such it should be used with caution. A warning will be
        generated if more than 50\% of either the presence or absence points will be deleted based on
        the ratio specified by the user. Background points are ignored by this module (they are read
        in and written out, but not assigned to either the test or training split).
\item Seed (optional)
        The random number seed used to split the data.  If one desires to reproduce results from a
        previous split fit, one must enter the random number seed that is reported in the console
        output from that split step.

\textbf{Default value} = Randomly selected

\textbf{Options}

\begin{itemize}
\item Any integer between -2147483647 and 2147483647
\end{itemize}

\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item outputMDS 
        This is an MDS file to be further used in the downstream workflow.  It is nearly identical
        to the input MDS file but with an added column under the header ``EvalSplit" for each non-
        background observation will be labeled either ``test", ``train", or ``NA" indicating whether
        the observation will be withheld for producing final modelevaluation metrics, training the
        model, or excluded from the analysis respectively.

\textbf{Common connections}

\begin{itemize}
\item This port can connect to either the CovariateCorrelationAndSelection, ModelSelectionSplit,
ModelSelectionCrossValidation, or directly to any of the R model modules (BoostedRegressionTree,
GLM, MARS, or RandomForest).
\item While it could technically also connect to the input port of ModelEvaluationSplit this would not
make sense for SAHM workflows.  The results of connecting to one of these modules has not been
tested and could cause errors or subtle but significant problems with subsequent modeling.
\end{itemize}

\end{itemize}
References:

       Hastie T, Tibshirani R, Friedman JH. 2009. The Elements of Statistical Learning: Data Mining,
        Inference, and Prediction. New York: Springer-Verlag. 744 pp. 2nd ed.

\section{ModelSelectionCrossValidation}


        The ModelSelectionCrossValidation module provides another tool for model selection by
    splitting the field data observations into cross validation folds.  This should not be used with
    the ModelSelectionSplit but can be used with the ModelEvaluationSplit in which case only the
    training portion of the ModelEvalutationSplit is partitioned into folds.  If specified then the
    individual models will fit a model using all of the data and report this as the training
    results.  Following the model fitting step sub-models with be fit to each set of n-1 folds and
    then evaluation metrics calculated on the remaining fold.  These will show up as ranges in the
    AUC plot, means and standard deviations are reported in textual output and box plots in across
    model comparison plots.  Evaluation metrics for each individual fold are reported in the across
    model comparison csv.  The cross validation method incorporated here was originally written for
    evaluation of MARS models by Leathwick et. al. 2006.  The current implementation does not
    attempt any sort of model averaging but rather is only used for calculation of evaluation
    metrics.  The ModelSelectionCrossValidation module makes better use of data then the
    ModelSelectionSplit as it uses all of the data to fit the final model but can be substantially
    more time consuming.

        Under most circumstances the cross validation evaluation metrics reported by this module do
    not indicate how the the model might perform if applied to an independent set of data but rather
    are to be used only for model selection purposes.  The first issue is that when cross validation
    is applied any feature selection based on the relationship between the response and the
    predictors must be carried out on each cross validation training set.  The
    CovariateCorrelationAndSelection module includes an exploration of the relationship between the
    predictors and the response and thus would need to be carried out for each for each cross
    validation training set.  The second issue is that it is invalid to use an evaluation metric for
    model selection and then report that metric for only the best performing model without
    acknowledgement to the total number of models that were considered and the range of the
    evaluation metrics.  This module ignores background points.


\subsection*{Input Ports}
\begin{itemize}
\item inputMDS (mandatory)
        This is the input data set consisting of locational data for each sample point, the values
        of each predictor variable at those points, and if established, a field denoting the weight
        that will be assigned to each point in modeling. This input is usually provided by the
        upstream steps that precede the Test Training Split module. Any value entered here (e.g.,
        specifying another existing MDS on the file system) will override the input specified by a
        model connection in the visual display.

\textbf{Common connections}

\begin{itemize}
\item This port can connect to the output port on MDSBuilder, ModelEvaluationSplit
\item While it could technically also connect to the output from another ModelSelectionCrossValidation,
CovariateCorrelationAndSelection, or ModelSelectionSplit this would not make sense for SAHM
workflows.  The results of connecting to one of these modules has not been tested and could cause
errors or subtle but significant problems with subsequent modeling.
\end{itemize}

\item nFolds (optional)
        The number of folds into which the data should be partitioned.  A trade-off exists in
        selecting the number of folds to use for cross validation.  When nFolds is close to the
        total number of observations the prediction error is nearly unbiased as the cross validation
        sample size is nearly equal to the total sample size but because the training sets are
        nearly identical in this case variance of the prediction error can be quite high (Hastie et.
        al 2009).

\textbf{Default value} = 10

\textbf{Options}

\begin{itemize}
\item Any integer less than the number of data points is valid but this is generally either set to 3 or 10
\end{itemize}

\item Stratify (optional)
        Indicate whether cross-validation folds should be stratified by the response

\textbf{Default value} = True (Checked)

\textbf{Options}

\begin{itemize}
\item True (Checked)
\item False (Unchecked)
\end{itemize}

\item Seed (optional)
        The random number seed used to split the data.  If one desires to reproduce results from a
        previous split fit, one must enter the random number seed that is reported in the console
        output from that split step.

\textbf{Default value} = Randomly selected

\textbf{Options}

\begin{itemize}
\item Any integer between -2147483647 and 2147483647
\end{itemize}

\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item outputMDS 
        This is an MDS file to be further used in the downstream workflow.  It is nearly identical
        to the input MDS file but with an added column under the header ``Split". for each non-
        background observation will be labeled either a number from 1 to the number of folds
        selected or ``NA" indicating which fold each observation has been partitioned into or whether
        it will not be included in the Model Selection step likely because it is being withheld for
        model evaluation.

\textbf{Common connections}

\begin{itemize}
\item This port can connect to either the CovariateCorrelationAndSelection or directly to any of the R
model modules (BoostedRegressionTree, GLM, MARS, or RandomForest).
\item While it could technically also connect to the input port of ModelEvaluationSplit,
ModelSelectionCrossValidation or ModelSelectionSplit this would not make sense for SAHM workflows.
The results of connecting to one of these modules has not been tested and could cause errors or
subtle but significant problems with subsequent modeling.
\end{itemize}

\end{itemize}
References:

    Hastie T, Tibshirani R, Friedman JH. 2009. The Elements of Statistical Learning: Data Mining,
        Inference, and Prediction. New York: Springer-Verlag. 744 pp. 2nd ed.

\section{CovariateCorrelationAndSelection}


        The CovariateCorrelationAndSelection view provides a breakpoint in the modeling workflow for
    the user to assess how well each variable explains the distribution of the sampled data points
    and to remove any variables that may exhibit high correlation with others.

        The display shows the n variables that have the highest total number of correlations above a
    threshold with other predictors using the maximum of the Pearson, Spearman and Kendall
    coefficient. The column heading over each variable displays the number of other variables with
    which the environmental predictor is correlated using the user supplied threshold which defaults
    to .7.  Radio buttons are available to limit the display and correlation calculations to any
    combination of presence, absence, or background points.  The first column in the plot shows the
    relationship between the response and each predictor.  Row labels indicate the maximum of the
    Spearman and Pearson correlation coefficient and a locally weighted smooth has been added to
    help distinguish the nature of the relationship.

        The remaining plots make up a square with histograms for each variable displayed on the
    diagonal.  Their respective graphical display and correlation with other variables can be found
    by locating the row/column intersection between each (above and below the diagonal).  The
    scatter plot along with a locally weight smooth is shown below the diagonal.  Presence records
    are represented by red points, absence by green, and background are yellow.  Above the diagonal
    is the correlation coefficient between the two predictors.  If Spearman or Kendall correlation
    coefficient is larger than the Pearson correlation coefficient then an s or k will show up in
    the bottom right corner of this box.

        A user is provided with the opportunity to select a new set of the environmental predictor
    variables and “Update” the Covariate Correlation screen to investigate the relationships among
    the new variables selected.  Variables with a high degree of correlation with other variables
    should generally be unchecked in their respective radio buttons, and will be excluded from
    subsequent analysis steps in the model workflow.

        Multiple iterations can be run at this screen, allowing the user to investigate the
    relationships among the environmental predictor variables and choose the most appropriate set to
    be used in the subsequent modeling. When the desired set of variables has been chosen, the “OK”
    button is selected and processing will resume in the VisTrails workflow.


\subsection*{Input Ports}
\begin{itemize}
\item inputMDS (mandatory)
        The file to select from.  If this file contains unselected layers (0 in the second header
        line) these will initially appear deselected in the GUI.

\textbf{Common connections}

\begin{itemize}
\item The inputMDS can come from any module that outputs an MDS file.  These are: MDSBuilder,
ModelEvaluationSplit, ModelSelectionSplit, and ModelSelectionCrossValidation.
\end{itemize}

\item selectionName (optional)
        This serves two purposes.  First to uniquely identify a given selection.  This unique name
        is used to determine if a selection has been previously made, to apply for example.  And
        secondly to provide something that can be changed to trigger VisTrails to rerun this module
        even if nothing upstream has changed.
\textbf{Options}

\begin{itemize}
\item Any text can be used.
\end{itemize}

\item ShowGUI (optional)
        This Boolean indicates whether to stop execution and display the GUI for user interaction.
        In some cases such as exploration you might want to make a selection in a previous run and
        then change this to false so that the selection will apply to subsequent runs without
        interrupting execution.

\textbf{Default value} = True

\textbf{Options}

\begin{itemize}
\item True - The GUI will be shown.
\item False - The GUI will not be shown, execution will not be interrupted, but the previous selection
made with the specified selectionName will be applied to the current MDS file.
\end{itemize}

\item numPlots (optional)
        The number of variables to display at a time in the plot frame.

\textbf{Default value} = 8

\textbf{Options}

\begin{itemize}
\item An integer greater than 1 and generally no greater than 12
\end{itemize}

\item minCor (optional)
        The minimum correlation used to summarize the number of other variables each variable is
        highly correlated with.

\textbf{Default value} = 0.7

\textbf{Options}

\begin{itemize}
\item A decimal number between 0 and 1.
\end{itemize}

\item corsWithHighest (optional)
        If one desires to view only other parameters that have a correlation above the specified
        threshold with the parameter than has the highest number of total correlations with other
        parameters then this should be set to true.  Otherwise, by default, the parameters that are
        selected for display will be the set of parameters that have the highest number of
        correlations with other parameters above the given threshold.

\textbf{Default value} = False

\textbf{Options}

\begin{itemize}
\item True (Checked)
\item False (Unchecked)
\end{itemize}

\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item outputMDS 
        This is the output MDS file with the user supplied selection applied.

\textbf{Common connections}

\begin{itemize}
\item The output from the CovariateCorrelationAndSelection will generally connect to one of the model
modules (BoostedRegressionTree, GLM, MARS, RandomForest, or Maxent)
\item If using Maxent the output from CovariateCorrelationAndSelection might also conect to the
RasterFormatConverter.
\end{itemize}

\end{itemize}


\section{ModelOutputType}



            This module is a required class for other modules and scripts within the
            SAHM package. It is not intended for direct use or incorporation into
            the VisTrails workflow by the user.


\subsection*{Input Ports}
\begin{itemize}
\item value (optional)
        NA
\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item value\_as\_string (optional)
        ToDo
\end{itemize}
References:

    NA

\section{ResponseType}



            This module is a required class for other modules and scripts within the
            SAHM package. It is not intended for direct use or incorporation into
            the VisTrails workflow by the user.


\subsection*{Input Ports}
\begin{itemize}
\item value (optional)
        NA
\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item value\_as\_string (optional)
        ToDo
\end{itemize}
References:

    NA

\section{ResampleMethod}



            This module is a required class for other modules and scripts within the
            SAHM package. It is not intended for direct use or incorporation into
            the VisTrails workflow by the user.


\subsection*{Input Ports}
\begin{itemize}
\item value (optional)
        NA
\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item value\_as\_string (optional)
        ToDo
\end{itemize}
References:

    NA

\section{AggregationMethod}



            This module is a required class for other modules and scripts within the
            SAHM package. It is not intended for direct use or incorporation into
            the VisTrails workflow by the user.


\subsection*{Input Ports}
\begin{itemize}
\item value (optional)
        NA
\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item value\_as\_string (optional)
        ToDo
\end{itemize}
References:

    NA

\section{MergedDataSet}



            This module is a required class for other modules and scripts within the
            SAHM package. It is not intended for direct use or incorporation into
            the VisTrails workflow by the user.


\subsection*{Input Ports}
\begin{itemize}
\item mdsFile (optional)
        NA
\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item value (optional)
        ToDo
\end{itemize}
References:

    NA

\section{PointAggregationMethod}



            This module is a required class for other modules and scripts within the
            SAHM package. It is not intended for direct use or incorporation into
            the VisTrails workflow by the user.


\subsection*{Input Ports}
\begin{itemize}
\item value (optional)
        NA
\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item value\_as\_string (optional)
        ToDo
\end{itemize}
References:

    NA

\section{Model}



            This module is a required class for other modules and scripts within the
            SAHM package. It is not intended for direct use or incorporation into
            the VisTrails workflow by the user.


\subsection*{Input Ports}
\begin{itemize}
\item mdsFile (optional)
        NA
\item makeBinMap (optional)
        NA
\item makeProbabilityMap (optional)
        NA
\item makeMESMap (optional)
        NA
\item ThresholdOptimizationMethod (optional)
        NA
\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item modelWorkspace (optional)
        ToDo
\item BinaryMap (optional)
        ToDo
\item ProbabilityMap (optional)
        ToDo
\item ResidualsMap (optional)
        ToDo
\item MessMap (optional)
        ToDo
\item MoDMap (optional)
        ToDo
\item modelEvalPlot (optional)
        ToDo
\item ResponseCurves (optional)
        ToDo
\item Text\_Output (optional)
        ToDo
\end{itemize}
References:

    NA

\section{PredictorList}


        This module is a required class for other modules and scripts within the SAHM package. It is
    not intended for direct use or incorporation into the VisTrails workflow by the user.


\subsection*{Input Ports}
\begin{itemize}
\item value (optional)
        NA
\item addPredictor (optional)
        NA
\end{itemize}

\subsection*{Output Ports}
\begin{itemize}
\item value (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\item value\_as\_string (optional)
        ToDo
\end{itemize}
References:

    NA

\end{document}