sohrabtowfighi / pysr

a pure-python symbolic regression library built on deap

pysr

a pure-python symbolic regression library built on deap

Installation

We use conda, and so should you.

1. Create new environment with: conda env create -n ENVIRONMENT_NAME -f environment.yml where ENVIRONMENT_NAME is the name you want for your new environment
2. Activate with source activate ENVIRONMENT_NAME on linux/OSX or activate ENVIRONENT_NAME on windows
3. Deactivate with source deactivate

Usage

python -m scoop pysr.py csvfile numgens popsize

The first n+1 columns of the CSV are x_0 through x_n. The last column is y.

Recommended parameters

I use popsize=500 and numgens very large so I can just kill it with C-c.

Pickling

After every generation, the current population is pickled to a file called pickle. To restart from a picklefile, just run pysr.py in the same directory as pickle. To restart from the beginning, delete or move pickle.

Plotting

Just run with numgens and popsize equal to 0 to plot out the current best fit.