Save and Reload Samples

Use SamplingResult.save() when you want to generate Sobol/Saltelli samples once, store them on disk, and reuse them later without re-running the sampling step.

Save, reload, and analyze

import gsax
from gsax.benchmarks.ishigami import PROBLEM, evaluate

sampling_result = gsax.sample(PROBLEM, n_samples=4096, seed=42)

# Save a file stem, not a full filename with extension.
sampling_result.save("runs/ishigami", format="csv")

restored = gsax.load("runs/ishigami", format="csv")
Y = evaluate(restored.samples)
result = gsax.analyze(restored, Y)

print("Loaded rows:", restored.n_total)
print("First SampleID values:")
print(restored.samples_df.head())
print(result.S1)
print(result.ST)

Files written to disk

For path="runs/ishigami" and format="csv", save() writes:

• runs/ishigami.csv with the unique sample matrix.
• runs/ishigami.json with the Problem definition, base_n, expanded_n_total, and related metadata.
• runs/ishigami.npz only when expanded_to_unique is not the identity map.

The .npz file is skipped for storage efficiency when the expanded Saltelli rows already map one-to-one to the unique rows.

Choosing a format

• csv is the safest default for interop and version control.
• txt is useful for plain numeric pipelines.
• pkl is compact inside Python-only workflows.
• xlsx requires openpyxl.
• parquet requires pyarrow.

Use the same format when calling gsax.load(). Format is not inferred from the metadata file.

Practical caveats

• Saving persists the metadata that gsax.analyze() needs; do not drop the .json file.
• Reloaded samples are still the unique rows, so your model evaluation step remains Y = model(restored.samples).
• If you need a human-readable audit trail, samples_df is easier to inspect than the raw NumPy array.

See also

• Basic Example for the minimal Sobol workflow.
• Bootstrap Confidence Intervals if you want uncertainty bounds on a reloaded design.
• Advanced Workflow for a larger custom model that keeps Sobol and HDMR runs side by side.