Supervised ML: Nearest Neighbor Classifier

Nearest Neighbor Classifier Script Description

Use this to brush up on Python syntax (I reccomend Google).

1. Process and graph a CSV file containing biomedical data that relates hemoglobin levels, glucose levels, and chronic kidney disease (CKD).
2. Create n number of random test cases.
3. Determine whether the case is likely to have CKD depending on the classification of the nearest point.

Bonus: graph a line connecting the test case to its nearest neighbor.

Nearest Neighbor Classifier Function Descriptions

Except for graphing, most of these can be one line functions.

• openCSVFile function takes in no arguments and parses/organizes data from a CSV file into a 2-D numpy array with the columns being: hemoglobin, glucose, classification and each row being a case.

• normalizeData function takes in a 2D numpy array and scales down the first and second columns to range from 0-1 and outputs a 2D array with the normalized data. *It may be easiest to create a class to work with the array.

• graphCSVFile void function takes in a 2D numpy array and graphs with: the first column (hemoglobin) as the x-axis and second column (glucose) as the y-axis. The third column (classification) is used to determine the color of the points on the graph.

• findDistance function is either takes in cartesian coordinates and uses a simple use of the distance formula to return the distance between the two points.

• createTestCase function creates two random test cases (hemoglobin and glucose) from 0-1 and creates/returns a new 1D array with the two points.

• nearestNeighborIndex function takes in a 2D array of cartesian points and a randomly generated test case. Returns the index of the closest point in the 2D array to the test case.
  Method 1: Use indices and the distance function to create an array of points and return the index of the closest point.
  Method 2: Use KDTree if you do this, you will not need to use the distance formula since this uses a spatial library Returns the index of the closest point in the 2D array to the test case.

• graphNearestNeighbor void function takes in a 2D numpy array (and a cartesian coordinate depending on createTestCase) and graphs the first column (hemoglobin) as the x-axis and the second column (glucose) as the y-axis. The third column (classification) determines the color of the points. A randomly generated test case is graphed as a distinct point with a line connecting it to the nearest neighbor whose classification it takes on. A legend is generated in a reasonable position.

---

Supervised ML: K Nearest Neighbor Classifier

K Nearest Neighbor Classifier Script Description

Try to figure out how to do this before looking at function descriptions.

1. Process and graph a CSV file containing biomedical data that relates hemoglobin levels, glucose levels, and chronic kidney disease (CKD).
2. Create n number of random test cases.
3. Determine whether the case is likely to have CKD depending on the mode of the classifications of the k number of nearest points.

K Nearest Neighbor Classifier Functions Descriptions

• kNearestNeighbor function inputs a numpy array, a random point, and an integer k and returns an array of length k which holds the indices of the k number of nearest points to the test case.
  Method 1: Use the findDistance function from NearestNeighborClassifier.
  Method 2: Use a KD Tree this is the same as for NearestNeighbor, except you will be doing a query of k number of points.

• kNNClass function takes in an 2 arrays. One of the data and the other is a list of indices. Can use various statistical calculations to find the value of the most common class and return that class.

• graphKNearestNeighbor void function takes in two 1D and one 2D numpy arrays to graph. One of the 1D arrays is a random testCase with its own distinct points. The other 1D array is used to circle the k number of points closest to the test case. The 2D array contains information parsed from the CSV column. The first column (hemoglobin) is graphed as the x-axis and the second column (glucose) as the y-axis. The third column (classification) determines the color of the points. A legend is generated in a reasonable position.

---

Unsupervised ML: K Means Clustering

K Means Clustering Script Description

This is the most difficult part! Try to think outside of the box when programming the functions.

1. Process and graph a CSV file containing biomedical data that relates hemoglobin levels, glucose levels, and chronic kidney disease (CKD).
2. Randomly generate up to 10 centroids without issue.
3. Assign each centroid will have a classification. The nearest centroid to a point will determine the point's classification (decide what to do if the distances are equal yourself).
4. Create random test cases until centroids stop moving (literally stop updating).
5. Determine whether each case is likely to have CKD depending on the classification of the nearest centroid.

Bonus: Animate the iteration process/ save it as a gif.

K Means ClusteringClassifier Functions Descriptions

• randomCentroids function takes in an integer number of clusters to be generated.
  OR asks for k number of integer clusters Outputs a 2D array filled with random values between 0-1. The first column represents hemoglobin and the second column represents glucose. There are k number of rows representing the number of centroids and the classification of each centroid (i.e.: row index = classification value).
  OR you can have a third column with the classification value.
  This can be a one line function

• assignCentroids function takes in an array of normalized x (hemoglobin) and y (glucose) values from the CSV file and the randomly generated array of centroids from randomCentroids.
  Method 1: Use shape/KD tree to determine closest point to each centroid. *This is more effective and can be a one line function. Hint: this is the opposite of the KDTree used in NearestNeighbor and kNearestNeighbor in that you want to find the centroid closest to each point instead of the point closest to a randomly generated point.
  Method 2: Using the findDistance function from NearestNeighborClassifier, points are assigned the same classification as the nearest centroid. A 2D array of the normalized data and its classification are returned.

• updateCentroids function inputs the 2D array of centroid locations and of classified and normalized CSV data. The average x (hemo) and y (gluc) positions of all data points for each classifications are found and an updated 2D array with these average cartesian points as the location for the new centroids is returned along with the original cartesian points.
  This function has two parts: finding average x and y values and then changing the values of the 2-D centroid array

• iterate void function inputs information and iterate the original information until centArr = upCentArr (i.e.: centroids do not move). This is the most difficult portion of the project. Hint: use recursion.

• graphClusters void function takes in a 1D and a 2D numpy array to graph. The 1D array of centroid locations and classifictions have distinct points on the graph. The 2D array graphs points of normalized CSV data and colors them the same color as their corresponding centroids. A legend is generated in a reasonable position.
  In order to use at least 10 colors, it may be easiest to use a color map here.

• dataAnalysis void function takes in the original parsed CSV classifications and the final classifications of the data based on K-means clustering (use of centroids) and compares the two to find false/true positives/negatives.

Example of k-means clustering (the data is not from our particular project) Source