genetic_algo

Why use this?

When you have many columns in a dataset and you want to find the subset of the most predictive.

Dimension reduction? OK, but why not PCA?

Principal Component Analysis is good for some situations but only gives you where there is variance - it doesn't cut through noise. Also, PCA doesn't tell you which columns are most predictive of your output

Retaining original columns, right... Why not brute force?

If you have 80 columns and you know you only want to keep 20, you would have to brute force over 3.5*10^18 combinations to figure out the optimal solution.

OK, so it cuts out a lot of unpromissing potential solutions. Doesn't that mean it's probabilistic?

Yes and that's a good thing. If you could execute one billion possible solutions per second, it would still take you 1 billion seconds, it would still take you over 100 years. This algorithm will likely converge on a solution in about an hour on a 2017 bottom of the line laptop.

Nice. But your code is a mess.

I'm cleaning it up.