| Authors: | Sander Canisius, Maarten van Gompel |
|---|---|
| Contact: | proycon@anaproy.nl |
| Web site: | https://github.com/proycon/python-timbl/ |
python-timbl is a Python extension module wrapping the full TiMBL C++ programming interface. With this module, all functionality exposed through the C++ interface is also available to Python scripts. Being able to access the API from Python greatly facilitates prototyping TiMBL-based applications.
This is the 2013 release by Maarten van Gompel, building on the 2006 release by Sander Canisius. For those used to the old library, there is one backwards-incompatible change, adapt your scripts to use import timblapi instead of import timbl, as the latter is now a higher-level interface.
Since 2020, this only supports Python 3, Python 2 support has been deprecated.
python-timbl is free software, distributed under the terms of the GNU General Public License. Please cite TiMBL in publication of research that uses TiMBL.
python-timbl is distributed as part of LaMachine (https://proycon.github.io/LaMachine), which significantly simplifies compilation and installation. The remainder of the instructions in this section refer to manual compilation and installation.
python-timbl depends on two external packages, which must have been built
and/or installed on your system in order to successfully build python-timbl.
The first is TiMBL itself; download its tarball from TiMBL's homepage and
follow the installation instructions, recent Ubuntu/Debian users will find
timbl in their distribution's package repository. In the remainder of this
section, it is assumed that $TIMBL_HEADERS points to the directory that
contains timbl/TimblAPI.h, and $TIMBL_LIBS the directory that has
contains the Timbl libraries. Note that Timbl itself depends on additional
dependencies.
The second prerequisite is Boost.Python, a library that facilitates writing Python extension modules in C++. Many Linux distributions come with prebuilt packages of Boost.Python. If so, install this package; on Ubuntu/Debian this can be done as follows:
$ sudo apt-get install libboost-python libboost-python-dev
If not, refer to the Boost installation instructions to build and install
Boost.Python manually. In the remainder of this section, let $BOOST_HEADERS
refer to the directory that contains the Boost header files, and
$BOOST_LIBS to the directory that contains the Boost library files. If you
installed Boost.Python with your distribution's package manager, these
directories are probably /usr/include and /usr/lib respectively.
If both prerequisites have been installed on your system, python-timbl can be obtained through github:
$ git clone git://github.com/proycon/python-timbl.git $ cd python-timbl
and can then be built and installed with the following command:
$ pip install .
python-timbl offers two interface to the timbl API. A low-level interface contained in the module timblapi, which is very much like the C++ library, and a high-level object oriented interface in the timbl module, which offers a TimblClassifier class.
The high-level interface features as TimblClassifier class which can be used for training and testing classifiers. An example is provided in example.py, parts of it will be discussed here.
After importing the necessary module, the classifier is instantiated by passing it an identifier which will be used as prefix used for all filenames written, and a string containing options just as you would pass them to Timbl:
import timbl
classifier = timbl.TimblClassifier("wsd-bank", "-a 0 -k 1" )
Normalization of theclass distribution is enabled by default (regardless of the -G option to Timbl), pass normalize=False to disable it.
Training instances can be added using the append(featurevector, classlabel) method:
classifier.append( (1,0,0), 'financial') classifier.append( (0,1,0), 'furniture') classifier.append( (0,0,1), 'geographic')
Subsequently, you invoke the actual training, note that at each step Timbl may output considerable details about what it is doing to standard error output:
classifier.train()
The results of this training is an instance base file, which you can save to file so you can load it again later:
classifier.save()
classifier = timbl.TimblClassifier("wsd-bank", "-a 0 -k 1" )
classifier.load()
The main advantage of the Python library is the fact that you can classify instances on the fly as follows, just pass a feature vector and optionally also a class label to classify(featurevector, classlabel):
classlabel, distribution, distance = classifier.classify( (1,0,0) )
You can also create a test file and test it all at once:
classifier = timbl.TimblClassifier("wsd-bank", "-a 0 -k 1" )
classifier.load()
classifier.addinstance("testfile", (1,0,0),'financial' ) #addinstance can be used to add instances to external files (use append() for training)
classifier.addinstance("testfile", (0,1,0),'furniture' )
classifier.addinstance("testfile", (0,0,1),'geograpic' )
classifier.addinstance("testfile", (1,1,0),'geograpic' ) #this one will be wrongly classified as financial & furniture
classifier.test("testfile")
print "Accuracy: ", classifier.getAccuracy()
If you are writing a multithreaded Python application (i.e. using the
threading module) and want to benefit from actual concurrency,
side-stepping Python's Global Interpreter Lock, add the parameter
threading=True when invoking the TimblClassifier constructor. Take
care to instantiate TimblClassifier before threading. You can then call
TimblClassifier.classify() from within your threads. Concurrency only
exists for this classify method.
If you do not set this option, everything will still work fine, but you won't benefit from actual concurrency due to Python's the Global Interpret Lock.
For documentation on the low level timblapi interface you can consult the TiMBL API guide. Although this document actually describes the C++ interface to TiMBL, the latter is similar enough to its Python binding for this document to be a useful reference for python-timbl as well. For most part, the Python TiMBL interface follows the C++ version closely. The differences are listed below.
Naming style
In the C++ interface, method names are in UpperCamelCase; for example, Classify, SetOptions, etc. In contrast, the Python interface uses lowerCamelCase: classify, setOptions, etc.
Method overloading TiMBL's Classify methods use the C++ method overloading feature to provide three different kinds of outputs. Method overloading is non-existant in Python though; therefore, python-timbl has three differently named methods to mirror the functionality of the overloaded Classify method. The mapping is as follows:
# bool TimblAPI::Classify(const std::string& Line,
# std::string& result);
#
def TimblAPI.classify(line) -> bool, result
#
# bool TimblAPI::Classify(const std::string& Line,
# std::string& result,
# double& distance);
#
def TimblAPI.classify2(line) -> bool, string, distance
#
# bool TimblAPI::Classify(const std::string& Line,
# std::string& result,
# std::string& Distrib,
# double& distance);
#
def TimblAPI.classify3(line, bool normalize=true,int requireddepth=0) -> bool, string, dictionary, distance
#Thread-safe version of the above, releases and reacquires Python's Global Interprer Lock
def TimblAPI.classify3safe(line, normalize, requireddepth=0) -> bool, string, dictionary, distance
Note that the classify3 function returned a string representation of the
distribution in versions of python-timbl prior to 2015.08.12, now it returns an
actual dictionary. When using classify3safe (the thread-safe version) ,
ensure you first call initthreads after instantiating timblapi, and
manually call the initthreading() method.
Python-only methods
Three TiMBL API methods print information to a standard C++ output stream object (ShowBestNeighbors, ShowOptions, ShowSettings, ShowSettings). In the Python interface, these methods will only work with Python (stream) objects that have a fileno method returning a valid file descriptor. Alternatively, three new methods are provided (bestNeighbo(u)rs, options, settings); these methods return the same information as a Python string object.
scikit-learn wrapper
A wrapper for use in scikit-learn has been added. It was designed for use in scikit-learn Pipeline objects. The wrapper is not finished and has to date only been tested on sparse data. Note that TiMBL does not work well with large amounts of features. It is suggested to reduce the amount of features to a number below 100 to keep system performance reasonable. Use on servers with large amounts of memory and processing cores advised.