forecast a series.

Paste a numeric series and (optionally) its seasonal period. TimeSeriesForecast fits Holt-Winters triple exponential smoothing — level, trend, and seasonal terms with α/β/γ chosen by minimizing in-sample error — decomposes the series, and forecasts forward. Crucially it reports an honest holdout backtest (MAE/RMSE/MAPE) against a naive last-value baseline: a forecast that does not beat naive adds nothing. Dependency-light real numpy, no GPU.