如何在Polars透视表中按数据类型批量执行fill_null操作

Absolutely! Polars has you covered here—you can absolutely perform type-based bulk fill_null operations, which mirrors the behavior of dplyr's old mutate_if function. No need to manually list out every column name anymore.

Your Original Approach (Manual Column Specification) #

First, let's recap your initial implementation where you had to target columns by name:

import polars as pl

scores = pl.DataFrame(
    {
        "zone": ["North", "North", "North", "South", "South", "East", "East", "East", "East"],
        "funding": ["yes", "yes", "no", "no", "no", "no", "no", "yes", "yes"],
        "score": [78, 39, 76, 56, 67, 89, 100, 55, 80],
    }
)

(
    scores.group_by("zone", "funding").len()
    .with_columns(
        pl.col("len").over("funding"),
        pl.format(
            "{}%",
            (pl.col("len") * 100 / pl.sum("len"))
            .over("funding")
            .round(2),
        ).alias("perc"),
    )
    .pivot("funding", index="zone")
    .with_columns(
        pl.col("len_yes").fill_null(0),
        pl.col("perc_yes").fill_null("0%")
    )
)

Optimized Solution (Type-Based Bulk Fill) #

Here's the streamlined version that uses Polars' type selection to batch-apply fill_null based on column data types:

(
    scores.group_by("zone", "funding").len()
    .with_columns(
        pl.col("len").over("funding"),
        pl.format(
            "{}%",
            (pl.col("len") * 100 / pl.sum("len"))
            .over("funding")
            .round(2),
        ).alias("perc"),
    )
    .pivot("funding", index="zone")
    .with_columns(
        pl.exclude(pl.String).fill_null(0),  # Fill all non-string columns with 0
        pl.col(pl.String).fill_null("0%")   # Fill all string columns with "0%"
    )
)

How This Works #

• pl.exclude(pl.String) selects every column that isn't a string type (in your case, the numeric len_* columns), then applies fill_null(0) to all of them at once.
• pl.col(pl.String) targets only string-type columns (your perc_* columns), applying fill_null("0%") to each.

This approach is way more scalable—if you end up with more columns down the line, you won't have to update your code to add new column names; Polars will automatically handle all columns matching the specified types.

Result (Same as Your Original Output) #

Running the optimized code gives you the exact same desired result:

shape: (3, 5)
┌───────┬────────┬─────────┬─────────┬──────────┐
│ zone  ┆ len_no ┆ len_yes ┆ perc_no ┆ perc_yes │
│ ---   ┆ ---    ┆ ---     ┆ ---     ┆ ---      │
│ str   ┆ u32    ┆ u32     ┆ str     ┆ str      │
╞═══════╪════════╪═════════╪═════════╪══════════╡
│ South ┆ 2      ┆ 0       ┆ 40.0%   ┆ 0%       │
│ North ┆ 1      ┆ 2       ┆ 20.0%   ┆ 50.0%    │
│ East  ┆ 2      ┆ 2       ┆ 40.0%   ┆ 50.0%    │
└───────┴────────┴─────────┴─────────┴──────────┘

内容的提问来源于stack exchange，提问作者Vincent