Sort list of datetime in Python polars dataframe

I have problems working with datetime

my dataset is like this:

And I wish to sort the dates inside the list on every row, and get the first and second datetime, because some rows has more than 2 values. But I can't do it on Polars.

I'm trying with :

filter_df.filter(
    pl.col("order_purchase_timestamp").arr.sort()
    )

but I got this error:
> ComputeError: Filter predicate must be of type Boolean, got: SchemaMisMatch(Owned("Series of dtype: List(Datetime(Microseconds, None)) != Boolean"))

You can use .with_columns() context to select & modify columns.

col = pl.col("order_purchase_timestamp")

df.with_columns(
    col.arr.sort(reverse=True).arr.slice(0, 2)
)   #        /                      /
    #      sort by date            take slice [0:2]

Solution that follows from the discussion in the comments

To perform some action under column based on condition, you can use pl.when -> then -> otherwise construct. pl.when() takes some boolean Series (condition). Read more here.

col = pl.col("order_purchase_timestamp")

# duration (difference) between 2 dates (in microseconds)
diff = pl.duration(microseconds=(col.arr.get(0) - col.arr.get(1)))

df.filter(col.arr.lengths() >= 2).with_columns(
    col.arr.sort(reverse=True).arr.slice(0, 2)
).with_columns(
    less_30 = pl.when(diff <= pl.duration(days=30))\
                .then(True).otherwise(False)
)

Instead of operating on lists - you could .explode() the column.

df = pl.DataFrame({"letters": [["f", "a", "m"], ["b"], ["e", "d", "c"]]})

shape: (3, 1)
┌─────────────────┐
│ letters         │
│ ---             │
│ list[str]       │
╞═════════════════╡
│ ["f", "a", "m"] │
│ ["b"]           │
│ ["e", "d", "c"] │
└─────────────────┘

.with_row_count() can be used as a "group id":

df.with_row_count("group").explode("letters").sort("letters")

shape: (7, 2)
┌───────┬─────────┐
│ group | letters │
│ ---   | ---     │
│ u32   | str     │
╞═══════╪═════════╡
│ 0     | a       │
│ 1     | b       │
│ 2     | c       │
│ 2     | d       │
│ 2     | e       │
│ 0     | f       │
│ 0     | m       │
└───────┴─────────┘

You could then filter based on the group count ("size"):

(df.with_row_count("group")
   .explode("letters")
   .sort("letters")
   .filter(pl.count().over("group") > 1))

shape: (6, 2)
┌───────┬─────────┐
│ group | letters │
│ ---   | ---     │
│ u32   | str     │
╞═══════╪═════════╡
│ 0     | a       │
│ 2     | c       │
│ 2     | d       │
│ 2     | e       │
│ 0     | f       │
│ 0     | m       │
└───────┴─────────┘

.groupby().agg() can be used to recreate the lists.

.head / .limit to keep only the first N results.

(df.with_row_count("group")
   .explode("letters")
   .sort("letters")
   .filter(pl.count().over("group") > 1)
   .groupby("group")
   .agg(pl.col("letters").head(2)))

shape: (2, 2)
┌───────┬────────────┐
│ group | letters    │
│ ---   | ---        │
│ u32   | list[str]  │
╞═══════╪════════════╡
│ 0     | ["a", "f"] │
│ 2     | ["c", "d"] │
└───────┴────────────┘