Import Existing Results into a Store

If you have already run experiments and have results in a data file, you can pre-populate a stet store so that those runs are skipped when you restart.

The situation

You have a file such as my_results.csv with columns for your parameters and results:

alpha,seed,mse
0.1,0,0.42
0.1,1,0.38
0.2,0,0.51

You now want to decorate your experiment function with @once and re-run your full sweep, but without repeating the work already done.

Step 1: Pre-populate the store

Use the backend directly to record each parameter combination present in your existing data:

import pandas as pd
from stet.backends import get_backend
from pathlib import Path

df = pd.read_csv("my_results.csv")

backend = get_backend(Path("_stet_store.csv"))
for _, row in df.iterrows():
    backend.record({"alpha": row["alpha"], "seed": row["seed"]})

Only include the parameter columns, not result columns like mse. The store tracks which parameter combinations have been run, not what they produced.

Step 2: Decorate and re-run

import stet

@stet.once(key=["alpha", "seed"])
def run_experiment(alpha, seed):
    ...

for alpha in [0.1, 0.2, 0.3]:
    for seed in range(5):
        run_experiment(alpha, seed)

Combinations already in the store are skipped; only the missing ones execute.

Matching parameter names and values

The parameter names used in backend.record() must match the names stet extracts from your function, which are the argument names unless you override them with key=. Values are stored as strings internally, so 0.1 and "0.1" are treated the same.

You can verify the store looks correct before re-running:

import stet

stet.status()

See Inspect a Store for more detail on reading store contents.