Changelog¶
All significant changes to this project are documented in this release file.
| Legend | |
|---|---|
| Feature | New feature |
| Enhancement | Improvement in existing functionality |
| API Change | Changes in the API |
| Fix | Bug fix |
0.2.0 Aug 3, 2026¶
Added
- Feature
ForecastingAssistant.compare()backtests several forecaster/estimator configurations with the same cross-validation strategy and returns a metric-rankedComparisonResultleaderboard. Each successful candidate is available incandidates, a name-keyed mapping ofBacktestResultordered best to worst, and the winning configuration is exposed asbest_name/best_candidatefor direct reuse. A matchingcompareCLI command reports the leaderboard and supports--candidates,--metric,--from-profile, and--output-code.
Changed
- API Change
ForecastingAssistant.ask()now accepts a singleresultparameter in place of the previousforecast_resultandbacktest_resultparameters. Update calls fromask(forecast_result=...)/ask(backtest_result=...)toask(result=...). AnyExplainableResultis accepted, includingForecastResult,BacktestResult, andComparisonResult.
Fixed
- API Change
ForecastingAssistant.ask()now treats a suppliedresultas the single source of truth. Previously an explicitprofileorplantook precedence over the result's own while the context and code still came from the result, so the returned artifacts could describe different states.data,target,date_column,series_id_column,profile,plan, andstepsare now ignored with anIgnoredArgumentWarningwhenresultis provided.
0.1.0 Jul 13, 2026¶
First public release. skforecast-ai wraps the skforecast engine in a deterministic, rule-based assistant that profiles the data, selects a model, evaluates it, and returns the forecast together with the exact, runnable script that produced it. An optional LLM layer explains the decisions without ever changing them.
Maturity
The underlying forecasting engine (skforecast) is mature and production-grade. The skforecast-ai assistant layer is at 0.1.0, so its public API may still change.
Added
ForecastingAssistant: the main entry point, covering the full workflow:profile(),plan(),refine_plan(),forecast()/forecast_code(),create_cv(),backtest()/backtest_code(), andask().- Typer-based CLI mirroring the programmatic API, with persistent configuration.