`pl.from_records(series_of_structs)` used to work (in v0.20.8 to v0.20.15), but now errors (in v0.20.16) about polars HOT 8 CLOSED

In the DeprecationWarning, and in future implementations when it's actually deprecated, it'd be awesome if the .struct.unnest() solution was suggested.

That is exactly how I've written it ;)

DeprecationWarning:
  `from_records` does not support Series input; unpack Struct records using `srs.struct.unnest()` instead

from polars.

Awesome comments, thanks a lot!

from polars.

Change of plans; we're going to continue allowing this, but I've integrated the unnest fast-path directly. I'd still recommend calling series.unnest() as it seems clearer to me, but now you won't be penalised if you don't ;)

from polars.

.struct.unnest() can be used as an alternative.

>>> series_of_structs.struct.unnest()
shape: (1, 3)
┌──────────────────┬───────────────────────────┬────────────┐
│ wellbore_uwi     ┆ licensee_company_brand_id ┆ license_id │
│ ---              ┆ ---                       ┆ ---        │
│ str              ┆ i64                       ┆ i64        │
╞══════════════════╪═══════════════════════════╪════════════╡
│ 102071407712W500 ┆ 178                       ┆ 16519701   │
└──────────────────┴───────────────────────────┴────────────┘

from polars.

I can fix, as it appears to be a regression, but this is a very odd way to achieve the desired result 😅 (While fixing I'll add a DeprecationWarning about use of Series here, as it is definitely not intended to be used that way).

Note that @cmdlineluser's approach will also be dramatically faster.

from polars.

In the DeprecationWarning, and in future implementations when it's actually deprecated, it'd be awesome if the .struct.unnest() solution was suggested.

from polars.

Good stuff, thanks! I think that's the best way!

from polars.

Change of plans (again): passing a Series to from_records will raise a TypeError.

This function was never intended to handle Series inputs. The docs clearly state that a sequence-of-sequences is expected. So the fact that a Series was supported and parsed into a DataFrame was in fact a bug.

For a catch-all that parses all types of data, you can use the DataFrame constructor. For your specific use case, if you want to turn your Struct Series into a DataFrame, you should use struct.unnest, as @cmdlineluser has pointed out.

from polars.