[Python] List of int and floats in Python Polars casting to object

Why does the column floats in the following example casts to object instead of list[f64]? How can I change it to list[f64]?

import polars as pl

dfLists = pl.DataFrame({
'ints':[ [0,1], [4,3,2]],
'floats':[ [0.0,1], [2,3]],
'strings':[ ["0","1"],["2","3"]]
})
dfLists


shape: (2, 3)
┌───────────┬──────────┬────────────┐
│ ints ┆ floats ┆ strings │
│ --- ┆ --- ┆ --- │
│ list[i64] ┆ object ┆ list[str] │
╞═══════════╪══════════╪════════════╡
│ [0, 1] ┆ [0.0, 1] ┆ ["0", "1"] │
│ [4, 3, 2] ┆ [2, 3] ┆ ["2", "3"] │
└───────────┴──────────┴────────────┘


Note: It seems this behavior may be new, based on this blog post (.html) where the same example casts to list[f64].

Edit: After updating polars to version 0.17.2 the column casts to list[f64] as expected.

I would still like to know if it is possible to change the dtypes of the elements of a list. For example, adapting the previous example:

import polars as pl

dfLists = pl.DataFrame({
'ints':[ [0,1], [4,3,2]],
'object':[ [0,"1"], [2,3]],
'strings':[ ["0","1"],["2","3"]]
})

dfLists.select(pl.col("strings").list.eval(pl.element().cast(pl.Int16)))
# works

dfLists.select(pl.col("object").list.eval(pl.element().cast(pl.Int16))) # error
# error:
# PanicException: not implemented for dtype Object("object")

Continue reading...