This is the readme file for Transfer Learning Capable Symbolic Regression (TLC-SR). TLC-SR trains a recurrent neural network to perform symbolic regression on multiple datasets.
Create the environmental $TLCSR_DATA. Data generated from the algorithm will be stored at this location.
I can run this on mac using python 3.7.4. These scripts require the following packages: pycma, numpy, pandas, networkx, and keras (with tensorflow backend). I am currently using the following versions of these packages:
| Package | Version |
|---|---|
| pycma | 2.7.0 |
| numpy | 1.17.1 |
| pandas | 0.25.1 |
| networkx | 2.3 |
| Keras | 2.3.1 |
| tensorflow | 1.15.0 |
Data generated during training will be stored in $TLCSR_DATA/experiment<exp>/ where <exp> is the experiment number which you will specify when running the algorithm. To train TLC-SR network run
python3 run_tlcsr.py <rep> <exp> --use_benchmarks --test_index <index> --use_kexpressions --simultaneous_targetswhere <rep> and <exp> are positive integers corresponding to the experiment number and the repetition number inside that experiment and <index> is the index to the list of target functions, which chooses the test function. These numbers will be included in output filenames and/or locations.
Similarly, to run the control experiment -- genetic programming (GP) with age-fitness pareto optimization (AFPO) -- use the following command
python3 run_tlcsr.py <rep> <exp> --use_benchmarks --test_index <index> --use_kexpressions --simultaneous_targets --genetic_programmingTo recreate my results exactly use <exp> = 25 and repeat for <rep> = 0 through <rep> = 29 and repeat again for <index> = 0 through <index> = 5.