[José Morales (jmoralez)]

What

validate_format now raises a ValueError when the id or time column contains null values (None, NaN, NaT or NA in pandas; nulls in polars). The check runs even when target_col=None.

Why

Null ids and times aren't supported, but nothing rejected them, so they either failed far from the cause or produced wrong results. For example, polars process_df with ids ['a', 'a', None, 'b'] silently returned an indptr that gave the null series a row from 'a' and 'a' the null row, while pandas raised TypeError: '<=' not supported between instances of 'str' and 'NoneType'. Since process_df, fill_gaps, pipeline, rectify and plot_series all call validate_format, they now fail early with a clear message.

🤖 Generated with Claude Code

[Copilot]

🟢 Approval recommended

The focused checks run before the target-column early return, with regression coverage for both dataframe backends and no unresolved findings.

No open issues 🎉

What changed in this PR

This PR adds early rejection of null identifiers and timestamps to shared dataframe validation, preventing downstream processing errors and incorrect results.

Changes:

• Reject null id and time values in pandas and polars, including when target_col=None.
• Add validation tests across supported column types and null representations.
• Add a process_df regression test for null identifiers in both backends.

File	Description
utilsforecast/​validation.py	Adds backend-specific null detection and id/time validation.
tests/​test_validation.py	Tests null rejection with and without target validation.
tests/​test_processing.py	Verifies process_df rejects null identifiers early.

_{Review effort: Balanced}

[Saul (nasaul)]

null_count() doesn't count NaN in polars, so a float id column with NaN passes here while the pandas branch (hasnans) rejects it:

validate_format(pl.DataFrame({'unique_id': [1.0, float('nan')], 'ds': [1, 2], 'y': [1., 2.]}))  # no error

Those ids then reach process_df/fill_gaps and get misgrouped. Something like s.null_count() > 0 or (s.dtype.is_float() and s.is_nan().any()) would match pandas. Adding pl.Series([1.0, float('nan')]) to test_validate_format_null_ids would cover it.

[José Morales (jmoralez)]

NaNs get grouped together in polars

[Copilot]

🟢 Approval recommended

The checks are correctly placed, preserve optional-target behavior, and have focused regression coverage with no unresolved findings.

No open issues 🎉

_{Review effort: Balanced}