There may be NULL values in y or X. Handle this like in Polars (https://github.com/azmyrajab/polars_ols/blob/c647fd297b75e099bac921fd4f27cd88e5febc5d/polars_ols/__init__.py#L48)
null_policy: Strategy for handling missing data, it can be:
- "ignore": does no null handling - use this option if nulls are already handled upstream.
- "zero": simply zero fills nulls in both targets & features prior to fitting.
- "drop": drops any rows (across targets or features) which have nulls prior to computing
coefficients. For non-moving-window models, it then masks associated predictions
with nulls. For moving-window models, coefficients are forward-filled.
- "drop_y_zero_x": Similar to "drop", but only rows with null targets are masked.
If features contain remaining nulls, it fills them with zeros.
This option allows extrapolation.
- "drop_zero": Masks rows with null targets or features similar in fitting coefficients.
similar to "drop". However, for predictions, it uses zero-filled features
dotted with those coefficients instead of masking with null.
This option allows extrapolation.
There may be NULL values in y or X. Handle this like in Polars (https://github.com/azmyrajab/polars_ols/blob/c647fd297b75e099bac921fd4f27cd88e5febc5d/polars_ols/__init__.py#L48)
null_policy: Strategy for handling missing data, it can be:coefficients. For non-moving-window models, it then masks associated predictions
with nulls. For moving-window models, coefficients are forward-filled.
If features contain remaining nulls, it fills them with zeros.
This option allows extrapolation.
similar to "drop". However, for predictions, it uses zero-filled features
dotted with those coefficients instead of masking with null.
This option allows extrapolation.