symbolic_experiments

Repository for symbolic regression/classification experiments. The experiment files are sc.py and sr.py, run on a SLURM cluster using sbatch sr.sh. The number of trials are specified using a job array index in the shell scripts, and SLURM will schedule these sequentially. Other scripts within the experiment directories are provided to collect and analyze model results.

Running locally -

Download data set from here:

Code will look for this file ('sym_data_4.csv') alongside repository (in same directory that symbolic_experiments is in).

Run the following line from the same directory as sc.py or sr.py, and scoop will automatically collect available processors to distribute model evaluations over. You can watch this happen in your system's task manager.

python -m scoop sc.py 1

The integer provided as an argument at the end (the job index in the cluster version) will be used to initialize random processes, do training-test splits, and label the trial folder and results files.

Dependencies and versions:

For experiments:

Required:
python 3.7.1
numpy 1.16.1
pandas 0.25.3
scipy 1.2.0
deap 1.2.2
scoop 0.7.1.1(as written, can be modified to run single processor)

Optional*(for immediate plotting of results):
matplotlib 3.1.0
pygraphviz 1.3
scikit-learn 0.20.2

*Remote installations of python may not have plotting, so can turn these off in sr.py and sc.py. All results are stored in a pickle within a directory labeled with the trial number.

For sensitivity testing, clustering, and most figures:

salib 1.3.8
scikit-learn 0.20.2
matplotlib 3.1.0
seaborn 0.9.0

For networks and maps:

pygraphviz 1.3
cartopy 0.17.0

Other data:

The code that extracts the IWFM water pumping and delivery estimates from C2VSim can be found here.

The code that extracts the land use data from the California Pesticide Use Reports can be found here.

The California County Ag Commissioner data (economic) can be found here.