GeneticAlgorithm

A genetic algorithm for selecting optimal feature combinations with sklearn supervised ML objects

See the docstring of GeneticAlgorithm.py for a brief explanation for how this works

To do

• Add plotting functionality for the progression of fitness
• Test with various datasets and sklearn classifiers/regressors
• Think of a more efficient way of parallelizing - at the moment new workers are activated every time a new generation is trained

Useage

• This should work with any Sklearn classifer or regression object. It can be used as in this simple example

First, make a classifier of your choice, and select hyperparameters. Note that the generic algorithm does not select hyperparameters, only features to train on.

• RF = RandomForestClassifier(n_estimators=1,min_samples_split=2,min_samples_leaf=1)

Make a genetic algorithm object. X and Y here are dataframes of the predictors (X) and target vector (Y).

• GA = GeneticAlgorithm(X,Y,RF,njobs=4)

Call GA.fit(). This will run the algorithm. It may take a long time!

• GA.fit()

An optimal combination of columns is selected and this can then be used in future developments. The selected features are assigned to the variable 'GA.feature_selection'