👋 how much work would it be to add

mode

(most common value) to aggregations?

how much work would it be to add

mode

(most common value)

Probably be quite involved, medium-hard difficulty. Mostly due to the fact that it would require an implementation for both local and distributed cases.

apply the same UDF to multiple columns with

map_groups

You can pass multiple columns to the UDF in map groups, and also return multiple columns. Example

import daft
import statistics

from daft.functions.struct import unnest


@daft.udf(
    return_dtype=daft.DataType.struct(
        {"mode1": daft.DataType.int64(), "mode2": daft.DataType.int64()}
    )
)
def mode(data1, data2):
    return [
        {
            "mode1": statistics.mode(data1),
            "mode2": statistics.mode(data2),
        }
    ]


df = daft.from_pydict(
    {
        "group": ["a", "a", "a", "b", "b", "b"],
        "data1": [1, 2, 2, 3, 3, 4],
        "data2": [1, 1, 2, 3, 4, 4],
    }
)

df = (
    df.groupby("group")
    .map_groups(mode(daft.col("data1"), daft.col("data2")).alias("mode"))
    .select("group", unnest(daft.col("mode")))
)
print(df.collect())

got it and makes sense that

mode

will be tedious to implement. I figured I could pass multiple columns, but wanted to see if I could avoid