This repo hosts an OOP implementation of a Genetic Algorithm trying to find the most important words in a corpus, in order to reduce the dimensionality of the Neural Network input vector.

genetic.py contains the code for it accompanied by the basic mutation method and some crossover, selection methods based on different teqnuiches

For the procedure of selection, suit your needs by picking between:

• cost roullete selection
• rank roullete selection
• tournament

For the procedure of crossover:

• single-point crossover
• multi-point crossover
• uniform crossover

The main program (main.py) calls the genetic algorithm constructor and applies it in order to to find the 1000 most important words in a corpus, based on multiple documents. This is done by creating a fitness function based on the TF-IDF metric. When the algorithm converges, it stores it's solution in order to remove the words from the documents that are not present in that (corpus) solution. Neural network model is now fed again with the reduced document sizes aiming for improved accuracy. Neural Network model built and training can be found in NN.py