python - 在多列上使用 Pandas df.where 会产生意外的 NaN 值

Question

给定数据框

import pandas as pd

df = pd.DataFrame({
    'transformed': ['left', 'right', 'left', 'right'],
    'left_f': [1, 2, 3, 4],
    'right_f': [10, 20, 30, 40],
    'left_t': [-1, -2, -3, -4],
    'right_t': [-10, -20, -30, -40],
})

我想创建两个新列，根据 transformed 的内容从 left_* 或 right_* 中选择:

df['transformed_f'] = df['right_f'].where(
    df['transformed'] == 'right',
    df['left_f']
)

df['transformed_t'] = df['right_t'].where(
    df['transformed'] == 'right',
    df['left_t']
)

我得到了预期的结果

df
#    transformed  left_f  right_f  left_t  right_t  transformed_f  transformed_t
# 0  left              1       10      -1      -10              1             -1
# 1  right             2       20      -2      -20             20            -20
# 2  left              3       30      -3      -30              3             -3
# 3  right             4       40      -4      -40             40            -40

但是，当我尝试在一个操作中执行此操作时，我得到了包含 NaN 值的意外结果

df[['transformed_f', 'transformed_t']] = df[['right_f', 'right_t']].where(
    df['transformed'] == 'right',
    df[['left_f', 'left_t']]
)

df
#    transformed  left_f  right_f  left_t  right_t  transformed_f  transformed_t
# 0  left              1       10      -1      -10            NaN            NaN
# 1  right             2       20      -2      -20           20.0          -20.0
# 2  left              3       30      -3      -30            NaN            NaN
# 3  right             4       40      -4      -40           40.0          -40.0

有没有办法同时在多个列上使用 df.where()？

Answer 1

你很接近，只需添加 .values 或 .to_numpy() 切片使其成为 NDarray:

根据文档:

other : scalar, NDFrame, or callable Entries where cond is False are replaced with corresponding value from other. If other is callable, it is computed on the NDFrame and should return scalar or NDFrame. The callable must not change input NDFrame (though pandas doesn’t check it).

因此，当您直接输入数据帧的切片时，索引(列名称)不匹配，因此它不会更新 df，当您传递 .values 时，它会忽略索引并添加值。

df[['transformed_f', 'transformed_t']]=(df[['right_f', 'right_t']].
                        where(df['transformed'] == 'right',df[['left_f', 'left_t']].values))
print(df)

---

  transformed  left_f  right_f  left_t  right_t  transformed_f  transformed_t
0        left       1       10      -1      -10              1             -1
1       right       2       20      -2      -20             20            -20
2        left       3       30      -3      -30              3             -3
3       right       4       40      -4      -40             40            -40