Diagnostics

polars_ts.diagnostics

Residual diagnostics for forecast model evaluation. Closes #57.

acf(df, target_col='y', max_lags=20, id_col='unique_id')

Compute autocorrelation function with confidence bands.

Parameters

df Input DataFrame. target_col Column to compute ACF on. max_lags Maximum number of lags. id_col Column identifying each time series.

Returns

pl.DataFrame DataFrame with columns [id_col, "lag", "acf", "ci_lower", "ci_upper"]. Confidence bands use the 95% level (±1.96/√n).

pacf(df, target_col='y', max_lags=20, id_col='unique_id')

Compute partial autocorrelation function via Durbin-Levinson.

Parameters

df Input DataFrame. target_col Column to compute PACF on. max_lags Maximum number of lags. id_col Column identifying each time series.

Returns

pl.DataFrame DataFrame with columns [id_col, "lag", "pacf", "ci_lower", "ci_upper"].

ljung_box(df, target_col='y', lags=None, id_col='unique_id')

Ljung-Box test for residual autocorrelation.

Parameters

df Input DataFrame (typically residuals). target_col Column to test. lags Lag values at which to compute the test. Defaults to [10, 20]. id_col Column identifying each time series.

Returns

pl.DataFrame DataFrame with columns [id_col, "lag", "q_stat", "p_value"].