[IliassSjm]

What

Adds an opt-in split_locations: bool = False keyword to sb.events(). When enabled, every known location column (location, pass_end_location, carry_end_location, shot_end_location, goalkeeper_end_location) gets flat float companions <col>_x, <col>_y, and <col>_z when a third coordinate exists (e.g. shots). Original list columns are preserved, and the default behaviour is unchanged.

Implemented as a standalone helper (helpers.split_location_cols) so it can also be applied to an existing dataframe after the fact.

Why

The list-valued location columns are the first thing most users have to work around:

• they are unhashable, so drop_duplicates() and groupby on locations raise;
• they block vectorised numpy/pandas operations (everyone ends up writing the same .apply(lambda loc: loc[0]));
• they don't round-trip through Parquet — pd.read_parquet returns numpy arrays where lists were written, which silently breaks downstream isinstance(loc, (list, tuple)) checks.

I hit the third one in a real project (cached World Cup events to Parquet, reloaded, and every coordinate extraction returned NaN). The helper handles list, tuple and ndarray inputs uniformly, so it also fixes dataframes that have already been through a Parquet cache.

Tests

tests/test_split_locations.py — six offline unit tests covering the x/y split, conditional z creation, NaN handling for missing/malformed values, the Parquet-roundtrip (ndarray) case, absent columns, and dtype/hashability of the output columns.

Happy to extend this to competition_events() and the README examples if you'd take it.

[IliassSjm]

Checking in on this — happy to rebase or adjust scope if there's interest. The helper is self-contained and the tests are offline, so it should merge cleanly. No rush!

[orca-security-us]

Orca Security Scan Summary

Status	Check	Issues by priority
Passed	Infrastructure as Code	0   0   0   0	View in Orca
Passed	Secrets	0   0   0   0	View in Orca