[Python] Group-By column in polars DataFrame inside with_columns

I have the following dataframe:

import polars as pl

df = pl.DataFrame({
'ID': [1, 1, 5, 5, 7, 7, 7],
'YEAR': [2025, 2025, 2023, 2024, 2020, 2021, 2021]
})
shape: (7, 2)
┌─────┬──────┐
│ ID ┆ YEAR │
│ --- ┆ --- │
│ i64 ┆ i64 │
╞═════╪══════╡
│ 1 ┆ 2025 │
│ 1 ┆ 2025 │
│ 5 ┆ 2023 │
│ 5 ┆ 2024 │
│ 7 ┆ 2020 │
│ 7 ┆ 2021 │
│ 7 ┆ 2021 │
└─────┴──────┘


Now I would like to get the unique number of years per ID, i.e.

shape: (7, 3)
┌─────┬──────┬──────────────┐
│ ID ┆ YEAR ┆ UNIQUE_YEARS │
│ --- ┆ --- ┆ --- │
│ i64 ┆ i64 ┆ u32 │
╞═════╪══════╪══════════════╡
│ 1 ┆ 2025 ┆ 1 │
│ 1 ┆ 2025 ┆ 1 │
│ 5 ┆ 2023 ┆ 2 │
│ 5 ┆ 2024 ┆ 2 │
│ 7 ┆ 2020 ┆ 2 │
│ 7 ┆ 2021 ┆ 2 │
│ 7 ┆ 2021 ┆ 2 │
└─────┴──────┴──────────────┘


So I tried df.with_columns(pl.col('YEAR').over('ID').alias('UNIQUE_YEARS')) but this gives the wrong result. So I came up with

df.join(df.group_by('ID').agg(pl.col('YEAR').unique().len().alias('UNIQUE_YEARS')), on='ID', how='left')


which does gives correct result! But it looks a bit clunky, and I wonder if there is a more natural way using with_columns and over?

Continue reading...