Compare and save parameter runs

View Markdown

Choose Backtest.optimize when you want the highest-scoring result. Choose runtime.optimize when you want a persistent report for every candidate and later ranking/filtering.

Best-run search

bt.optimize(length=[10, 20, 40], maximize="Return [%]", return_heatmap=True) returns (best_result, scores). Candidate keys are Python field names. Install the optimize extra; see the complete example.

Saved experiment

Run the complete saved-experiment script. It declares inputs, supplies a search keyed by display titles, writes each Portfolio to SQLite, and queries the results.

1. Choose Values([...]) or Range(start, stop, step) for each title.
2. Supply the loaded strategy, data/context/instrument, a new local output path, and source/dependency hashes.
3. Call runtime.optimize, then inspect the returned ExperimentResult.
4. Reopen with open_result(path) for read-only queries without the original strategy.

Use fields() to discover sortable metrics and units. query returns selected scalars; load_portfolio(trial_id) returns the full report. Sorting does not rerun the strategy. Use backup(destination) while a run is writing rather than copying only its database file.

Resume with runtime.resume_optimization and the original experiment ID plus matching inputs, source, data, and settings. Changed material needs a new experiment. Saved experiments retain Portfolios, not plot/log history.

Parameters, queries, and resume · Portfolio report