Quickstart

Installation

Install the latest stable release:

pip install symbolic-regression-toolkit

Or install directly from the repository (recommended for the latest features):

pip install git+https://github.com/smeznar/SymbolicRegressionToolkit

For approaches that require PyTorch and pymoo (EDHiE):

pip install 'symbolic-regression-toolkit[approaches]'

Your first benchmark run

The snippet below loads two datasets from the Feynman benchmark, runs both ProGED and EDHiE on them, and prints progress.

import numpy as np
from SRToolkit.dataset import Feynman
from SRToolkit.approaches import EDHiE, ProGED
from SRToolkit.experiments import ExperimentGrid

# Load two 2-variable datasets from the Feynman benchmark
bm = Feynman()
ds_names = bm.list_datasets(num_variables=2, verbose=False)
datasets = [bm.create_dataset(ds_names[0]), bm.create_dataset(ds_names[1])]

# EDHiE trains a VAE once per symbol space; reuse the same weights for both datasets
# since they share the same number of variables
adapted_states = {
    "EDHiE": {ds_names[0]: "state_2v.pt", ds_names[1]: "state_2v.pt"}
}

grid = ExperimentGrid(
    datasets=datasets,
    approaches=[ProGED(), EDHiE()],
    num_experiments=3,
    results_dir="results/quickstart",
    adapted_states=adapted_states,
)

# Adapt any models whose state files are missing
grid.adapt_if_missing()

# Run all experiments sequentially
for job in grid.create_jobs():
    job.run()

# Print a completion table
grid.progress()

For larger runs, replace the for loop with grid.save_commands("commands.sh") and execute the file in parallel — see the Running Experiments guide.