stet

stet is a Python library that solves a common problem in computational research: when a long-running experiment script is re-run, whether after a crash, a time limit, or deliberately to extend a sweep, it should automatically skip any experiments that have already been completed.

Stet is a Latin proofreading instruction meaning "let it stand" - written beside a correction that should be ignored. When stet sees a parameter combination it has already run, it does the same: leave it, it's done.

The Problem

You're running a parameter sweep over thousands of combinations. Your script stops. Maybe it crashed, maybe you killed it, maybe you just want to add more parameter values. You restart it and it starts over from the beginning, re-running everything.

The Solution

Decorate your experiment function with @once:

import stet
import numpy as np

@stet.once(store='markov_runs.csv', key=['alpha', 'n_states', 'seed'])
def solve_markov(alpha, n_states, seed, n_iter=1000):
    # expensive computation here
    result = run_chain(alpha, n_states, seed, n_iter)
    # write your own output
    with open('markov_output.csv', 'a') as f:
        f.write(f"{alpha},{n_states},{seed},{result}\n")

for alpha in np.linspace(0.01, 1.0, 50):
    for n_states in [10, 50, 100]:
        for seed in range(20):
            solve_markov(alpha=alpha, n_states=n_states, seed=seed)

On restart, any already-completed (alpha, n_states, seed) combinations are skipped automatically.

Installation

$ uv add stet
$ uv add stet[parquet]  # with parquet support

or

$ python -m pip install stet
$ python -m pip install stet[parquet]  # with parquet support

Quick Start

import stet

@stet.once(store='_stet_store.csv', key=['alpha', 'beta'])
def run_experiment(alpha, beta, n_steps=1000):
    # your computation here
    pass

See the Getting Started tutorial for a complete walkthrough.