Caveat Emptor! There are some embarrassing errors that I need to fix, however, I think a look at the punch card will provide a complete explanation for their existance.

Three classifiers:

1. Bayesian classifier based on maximum-likelihood estimation
2. Bayesian classifer based on Parzen window estimation
3. Basic k-nearest neighbor rule

Included in this repo are the three different data sets that were used for testing. (corresponding credits and information are in their respective ./bin/*_readme.txt files):

• Iris
• UCI Wine
• Handwritten Digits

Simplified by using the run.py script, e.g.: python3 run.py classifier_name path_to_training_data [-h] [-t PATH_TO_TESTING_DATA] [-c PATH_TO_CLASSIFICATION_DATA] [-v]

The possible values for classifier_name are:

• mle for the MLE Bayesian classifier.
• parzen (or p) for the Parzen window Bayesian classifer.
• knn for the k-nearest neighbors classifier.

So, if you wanted to use the maximum likelihood classifer on the iris data set, then the command would be python3 run.py mle ./bin/iris_training.txt -t ./bin/iris_test.txt.

The training and testing files should have each instance on a separate line, with components separated by spaces. Per the following example:

class_number x0 x1 x2 x3
class_number x0 x1 x2 x3

For the classification data, the file should not include the class_number (e.g. each instance is separated by a new line).

To customize use of the classf module (e.g. make a custom run.py script), the module has a run command that can help, and following example demonstrates usage:

def myCustomFileParser(filepath):
    # parse file into a numpy array, with each instance a row in this array.
    # e.g. data[0] corresponds to the first instance's feature vector.

import classf
classifier = 'parzen'
training_data = './training.txt'
testing_data = './testing.txt'
verbose = True

classifier = classf.run(classifier, training_data, testing_data, verbose, myCustomFileParser)

###References, etc

• Sebastian Raschka's resources were a huge help in understanding.