groupby for list and struct type columns

[peterlietz]

Thank you for this absolutely wonderful library!

I'm afraid I hit a snag. What I tried to do was to group by a nested data type, as in:

df = pl.DataFrame({"a": [1, 2, 3], "b": [[1, 3, 4], [2, 4, 6], [17]]})
df.groupby("b").agg(pl.sum("a"))

This results in a not implemented panic.

I'm curious as to whether this is simply not implemented yet or whether this would contradict the underlying philosophy of polars.

Best regards Peter

[ritchie46]

We do not support grouping by a column of type list. I think we should improve the error message on that.

[peterlietz]

Thank you very much for the quick answer!

[pepelovesvim]

We do not support grouping by a column of type list. I think we should improve the error message on that.

@ritchie46 what do you think should be the error that comes out? DataTypeMisMatch?

[ritchie46]

I think a ComputeError would be most consistent.

For structs we could temporarily unnest -> do the groupby -> and nest again.

[peterlietz]

Just in case anybody else stumbles upon this, the workaround I am now using is to convert to "str". Not ideal, but does the trick.

df = pl.DataFrame({"a": [1, 2, 3], "b": [[1, 3, 4], [2, 4, 6], [17]]})
df = df.with_column(pl.col("b").arr.eval(pl.element().cast(pl.Utf8)).arr.join("|"))
df.groupby("b").agg(pl.sum("a"))