month_delta return incorrect value if passed a date object

[pounde]

I am attempting to replicate the functionality of relativedelta. The month_delta meets my current purposes but it return incorrect values if passed a date object.

This provided incorrect results. I would expect correct values or a ValueError

    df_ = df.filter(pl.col("date") <= end_date).with_columns(
        yr_for_calc=xdt.month_delta(pl.col("date"), end_date) / 12
    )

The following was my workaround:

    df_ = (
        df.filter(pl.col("date") <= end_date)
        .with_columns(end_date=end_date)
        .with_columns(
            yr_for_calc=xdt.month_delta(pl.col("date"), pl.col("end_date")) / 12
        )
    ).drop("end_date")

[MarcoGorelli]

thanks @pounde for the report

could you include a reproducible example please?

If I run

# ruff: noqa
from datetime import date, datetime
import polars as pl
import polars_xdt as xdt
df = pl.DataFrame(
    {
        "start_date": [
            date(2024, 3, 1),
            date(2024, 3, 31),
            date(2022, 2, 28),
            date(2023, 1, 31),
            date(2019, 12, 31),
        ],
        "end_date": [
            date(2023, 2, 28),
            date(2023, 2, 28),
            date(2023, 2, 28),
            date(2023, 1, 31),
            date(2023, 1, 1),
        ],
    },
)
print(df.with_columns(
    xdt.month_delta("start_date", date(2023, 2, 28)).alias("month_delta")
))
print(df.with_columns(
    xdt.month_delta("start_date", datetime(2023, 2, 28)).alias("month_delta")
))

then I do indeed get an error

polars.exceptions.ComputeError: the plugin failed with message: polars_xdt.month_delta only works on Date type. Please cast to Date first.

[pounde]

Apologies for not doing that from the start. I can't seem to replicate my code, nor can I replicate the code you posted. It is also worth noting that the object passed in the original issue was a 'date' object, not 'datetime'. I have updated the title to reflect that.

While attempting to run the code you provided, I get:

[line 22](vscode-notebook-cell:?execution_count=36&line=22)InvalidOperationError: Series month_delta, length 1 doesn't match the DataFrame height of 5

If you want expression: col("start_date")./usr/local/lib/python3.10/site-packages/polars_xdt/_internal.abi3.so:month_delta([2023-02-28]) to be broadcasted, ensure it is a scalar (for instance by adding '.first()').

Not sure why I didn't get this before, nor why it can't be broadcast.

[MarcoGorelli]

While attempting to run the code you provided, I get:

are you sure you got that while running the code I posted? could you post the full traceback please?

[pounde]

Apologies for the long tail. I created a fresh env:

• devcontainer
• Python 3.11.10
• Polars 1.14.0
• Polars-xdt 0.16.0

I still get the shape error. The stacktrace is:

{
	"name": "InvalidOperationError",
	"message": "Series month_delta, length 1 doesn't match the DataFrame height of 5

If you want expression: col(\"start_date\")./usr/local/lib/python3.11/site-packages/polars_xdt/_internal.abi3.so:month_delta([2023-02-28]) to be broadcasted, ensure it is a scalar (for instance by adding '.first()').",
	"stack": "---------------------------------------------------------------------------
InvalidOperationError                     Traceback (most recent call last)
Cell In[18], line 23
      4 import polars_xdt as xdt
      5 df = pl.DataFrame(
      6     {
      7         \"start_date\": [
   (...)
     21     },
     22 )
---> 23 print(df.with_columns(
     24     xdt.month_delta(\"start_date\", date(2023, 2, 28)).alias(\"month_delta\")
     25 ))
     26 print(df.with_columns(
     27     xdt.month_delta(\"start_date\", datetime(2023, 2, 28)).alias(\"month_delta\")
     28 ))

File /usr/local/lib/python3.11/site-packages/polars/dataframe/frame.py:9202, in DataFrame.with_columns(self, *exprs, **named_exprs)
   9056 def with_columns(
   9057     self,
   9058     *exprs: IntoExpr | Iterable[IntoExpr],
   9059     **named_exprs: IntoExpr,
   9060 ) -> DataFrame:
   9061     \"\"\"
   9062     Add columns to this DataFrame.
   9063 
   (...)
   9200     └─────┴──────┴─────────────┘
   9201     \"\"\"
-> 9202     return self.lazy().with_columns(*exprs, **named_exprs).collect(_eager=True)

File /usr/local/lib/python3.11/site-packages/polars/lazyframe/frame.py:2029, in LazyFrame.collect(self, type_coercion, predicate_pushdown, projection_pushdown, simplify_expression, slice_pushdown, comm_subplan_elim, comm_subexpr_elim, cluster_with_columns, collapse_joins, no_optimization, streaming, engine, background, _eager, **_kwargs)
   2027 # Only for testing purposes
   2028 callback = _kwargs.get(\"post_opt_callback\", callback)
-> 2029 return wrap_df(ldf.collect(callback))

InvalidOperationError: Series month_delta, length 1 doesn't match the DataFrame height of 5

If you want expression: col(\"start_date\")./usr/local/lib/python3.11/site-packages/polars_xdt/_internal.abi3.so:month_delta([2023-02-28]) to be broadcasted, ensure it is a scalar (for instance by adding '.first()')."
}

Thus far, I'm unable to replicate the issue I had so it may well have been user error. Let me know if there is more I can provide to help.

[MarcoGorelli]

Thanks

Sure, that should probably broadcast - @akmalsoliev fancy taking this on, as you'd introduced the feature?

[MarcoGorelli]

So, the issue is:

from datetime import date, datetime
import polars as pl
import polars_xdt as xdt
df = pl.DataFrame(
    {
        "start_date": [
            date(2024, 3, 1),
            date(2024, 3, 31),
            date(2022, 2, 28),
            date(2023, 1, 31),
            date(2019, 12, 31),
        ],
        "end_date": [
            date(2023, 2, 28),
            date(2023, 2, 28),
            date(2023, 2, 28),
            date(2023, 1, 31),
            date(2023, 1, 1),
        ],
    },
)
print(df.with_columns(
    xdt.month_delta("start_date", date(2023, 2, 28)).alias("month_delta")
))

outputs:

InvalidOperationError: Series month_delta, length 1 doesn't match the DataFrame height of 5

If you want expression: col("start_date")./home/marcogorelli/scratch/.venv/lib/python3.12/site-packages/polars_xdt/_internal.abi3.so:month_delta([2023-02-28]) to be broadcasted, ensure it is a scalar (for instance by adding '.first()').

[akmalsoliev]

So, the issue is:

from datetime import date, datetime
import polars as pl
import polars_xdt as xdt
df = pl.DataFrame(
    {
        "start_date": [
            date(2024, 3, 1),
            date(2024, 3, 31),
            date(2022, 2, 28),
            date(2023, 1, 31),
            date(2019, 12, 31),
        ],
        "end_date": [
            date(2023, 2, 28),
            date(2023, 2, 28),
            date(2023, 2, 28),
            date(2023, 1, 31),
            date(2023, 1, 1),
        ],
    },
)
print(df.with_columns(
    xdt.month_delta("start_date", date(2023, 2, 28)).alias("month_delta")
))

outputs:

InvalidOperationError: Series month_delta, length 1 doesn't match the DataFrame height of 5

If you want expression: col("start_date")./home/marcogorelli/scratch/.venv/lib/python3.12/site-packages/polars_xdt/_internal.abi3.so:month_delta([2023-02-28]) to be broadcasted, ensure it is a scalar (for instance by adding '.first()').

Hey, interesting issue, this shouldn't happen, there should be an error raised that date(2023, 2, 28) is not a Series or if to be included then should be converted into Series, don't know if we should go that path, but do see utility in it.

polars-xdt/src/month_delta.rs

Line 120 in aa2e214

pub(crate) fn impl_month_delta(start_dates: &Series, end_dates: &Series) -> PolarsResult<Series> {

this works fine

from datetime import date

import polars as pl

import polars_xdt as xdt

df = pl.DataFrame(
    {
        "start_date": [
            date(2024, 3, 1),
            date(2024, 3, 31),
            date(2022, 2, 28),
            date(2023, 1, 31),
            date(2019, 12, 31),
        ],
        "end_date": [
            date(2023, 2, 28),
            date(2023, 2, 28),
            date(2023, 2, 28),
            date(2023, 1, 31),
            date(2023, 1, 1),
        ],
    },
)

df = df.with_columns(test=date(2023, 2, 28)).with_columns(
    result=xdt.month_delta("start_date", "test")
)

print(df)

out:

shape: (5, 4)
┌────────────┬────────────┬────────────┬────────┐
│ start_date ┆ end_date   ┆ test       ┆ result │
│ ---        ┆ ---        ┆ ---        ┆ ---    │
│ date       ┆ date       ┆ date       ┆ i32    │
╞════════════╪════════════╪════════════╪════════╡
│ 2024-03-01 ┆ 2023-02-28 ┆ 2023-02-28 ┆ -12    │
│ 2024-03-31 ┆ 2023-02-28 ┆ 2023-02-28 ┆ -13    │
│ 2022-02-28 ┆ 2023-02-28 ┆ 2023-02-28 ┆ 12     │
│ 2023-01-31 ┆ 2023-01-31 ┆ 2023-02-28 ┆ 1      │
│ 2019-12-31 ┆ 2023-01-01 ┆ 2023-02-28 ┆ 38     │
└────────────┴────────────┴────────────┴────────┘