I’m trying to reproduce pandas.cut binning using polars.Series.cut.

MRE of values and breakpoints:

scores = [1111, 65, 88, -1111, 92]
breaks = [0, 50, 60, 70, 80, 90, 100]

With pandas.cut, the bin is null if the value is outside the defined edges:

df = pd.DataFrame({'score': scores})
df['bin'] = pd.cut(df['score'], breaks)

#    score            bin
# 0   1111            NaN  <- null in pandas
# 1     65   (60.0, 70.0]
# 2     88   (80.0, 90.0]
# 3  -1111            NaN  <- null in pandas
# 4     92  (90.0, 100.0]

But with polars.Series.cut, it seems we’re forced to include the inf and -inf bins:

df = pl.DataFrame({'score': scores})
df.with_columns(bin=pl.col('score').cut(breaks))

# shape: (5, 2)
# ┌───────┬────────────┐
# │ score ┆ bin        │
# │ ---   ┆ ---        │
# │ i64   ┆ cat        │
# ╞═══════╪════════════╡
# │ 1111  ┆ (100, inf] │  <- not null in polars
# │ 65    ┆ (60, 70]   │
# │ 88    ┆ (80, 90]   │
# │ -1111 ┆ (-inf, 0]  │  <- not null in polars
# │ 92    ┆ (90, 100]  │
# └───────┴────────────┘

I wasn’t able to set them to np.nan manually either:

df.with_columns(
    bin=pl.when(pl.col('bin').cast(pl.Utf8).str.contains('inf'))
          .then(np.nan)
          .otherwise(pl.col('bin'))
)

# ComputeError: failed to determine supertype of f64 and cat

How can we replicate the pandas.cut bins in Polars?