/pattern-recognition

Pattern Recognition methods, including Bayes, Linear and Quadratic Classifiers, Sequential Tests, Clustering Methods (Kmeans and Square Error).

Primary LanguageJupyter NotebookMIT LicenseMIT

Pattern Recognition

Pattern Recognition methods, including:

  • Bayes, Linear, Quadratic Classifier
  • Sequential Tests
  • Clustering (Kmeans and Square Error)

Additionally, not-MNIST model for Serbian Vowels is implemented.

What can these methods do?

Some of the classification results are shown below. Various classifiers are implemented, including Linear, Bayes and Quadratic.
Bayes Classifier Quadratic Classifier

Wald sequential test is capable of classifying examples, even if the classes are totally non-separable, as shown below.
PDFs for two classes Wald classification illustrated

The result of Kmeans clustering method is shown below. Additionally, you can find and plot stats for the number of steps the method in average needs to converge.
Kmeans result Stats for Kmeans

What is not-MNIST for Serbian Vowels?

Serbian language has 5 vowels, which are shown below:

A E I O U

Classification model is implemented in Keras, using dataset of 600 images.

How to run the tests?

To run any test simply go to the directory above 'source' and type the following command in your terminal.

python -m source.test.test_script

Test script can be any from the directory 'test':

  • test_bayes - To test Bayes Classifier
  • test_wald - To test Wald Sequential Test
  • test_linear - To test Linear Classifier
  • test_quadratic - To test Quadratic Classifier
  • test_kmeans - To test Kmeans Clustering
  • test_square_clustering - To test Square Error Clustering