Pandas数据框内插值使用常数值。

huangapple
go评论107阅读模式
英文:

Pandas Dataframe interpolate inside with constant value

问题

如何实现：

  1. [In1]:  df = pd.DataFrame({
  2.             'col1': [100, np.nan, np.nan, 100, np.nan, np.nan, np.nan],
  3.             'col2': [np.nan, 100, np.nan, np.nan, np.nan, 100, np.nan]})
  4.         df

结果为:

  1. [Out1]:       col1    col2
  2.         0      100     NaN
  3.         1      NaN     100
  4.         2      NaN     NaN
  5.         3      100     NaN
  6.         4      NaN     NaN
  7.         5      NaN     100
  8.         6      NaN     NaN

转化为:

  1. [Out2]:       col1    col2
  2.         0      100     NaN
  3.         1        0     100
  4.         2        0       0
  5.         3      100       0
  6.         4      NaN     NaN
  7.         5      NaN     100
  8.         6      NaN     NaN

所以基本上我想要在内部区域进行插值/填充NaN，使用limit=2。请注意，在col2中有三个连续的NaN，但只有其中两个被替换为零。

英文:

How to make:

  1. [In1]:  df = pd.DataFrame({
  2.             'col1': [100, np.nan, np.nan, 100, np.nan, np.nan, np.nan],
  3.             'col2': [np.nan, 100, np.nan, np.nan, np.nan, 100, np.nan]})
  4.         df
  6. [Out1]:       col1    col2
  7.         0      100     NaN
  8.         1      NaN     100
  9.         2      NaN     NaN
  10.         3      100     NaN
  11.         4      NaN     NaN
  12.         5      NaN     100
  13.         6      NaN     NaN

into:

  1. [Out2]:       col1    col2
  2.         0      100     NaN
  3.         1        0     100
  4.         2        0       0
  5.         3      100       0
  6.         4      NaN     NaN
  7.         5      NaN     100
  8.         6      NaN     NaN

So basically I want to interpolate/fill NaN's with zero only for the inside area and a limit=2. Note in col2 there are three consecutive NaN's in the middle and only two of them are replaced with zero.

答案1

得分: 1

以下是翻译好的部分：

你可以构建掩码来识别非-NAs，以及内部的值（借助双重 cummax 函数）：

  1. m = df.notna()
  2. m2 = m.cummax() & m[::-1].cummax()
  4. out = df.fillna(df.mask(m, 0).ffill(limit=2).where(m2))

或者使用 interpolate 函数：

  1. m = df.notna()
  3. out = df.fillna(df.mask(m, 0).interpolate(limit=2, limit_area='inside'))
  5. # 或者如果只有数字
  6. out = df.fillna(df.mul(0).interpolate(limit=2, limit_area='inside'))

输出结果：

  1.     col1   col2
  2. 0  100.0    NaN
  3. 1    0.0  100.0
  4. 2    0.0    0.0
  5. 3  100.0    0.0
  6. 4    NaN    NaN
  7. 5    NaN  100.0
  8. 6    NaN    NaN
英文:

You can build masks to identify the non-NAs, and the inner values (with help of a double cummax):

  1. m = df.notna()
  2. m2 = m.cummax() & m[::-1].cummax()
  4. out = df.fillna(df.mask(m, 0).ffill(limit=2).where(m2))

Or with interpolate:

  1. m = df.notna()
  3. out = df.fillna(df.mask(m, 0).interpolate(limit=2, limit_area='inside'))
  5. # or if you only have numbers
  6. out = df.fillna(df.mul(0).interpolate(limit=2, limit_area='inside'))

Output:

  1.     col1   col2
  2. 0  100.0    NaN
  3. 1    0.0  100.0
  4. 2    0.0    0.0
  5. 3  100.0    0.0
  6. 4    NaN    NaN
  7. 5    NaN  100.0
  8. 6    NaN    NaN

答案2

得分: 0

我们可以这样做：

  1. out = df.ffill(limit=2).mask(df.bfill().isna())
  2. out = out.mask(out.ne(df) & out.notna(), 0)
  3. Out[83]:
  4.     col1   col2
  5. 0  100.0    NaN
  6. 1    0.0  100.0
  7. 2    0.0    0.0
  8. 3  100.0    0.0
  9. 4    NaN    NaN
  10. 5    NaN  100.0
  11. 6    NaN    NaN
英文:

We could do

  1. out = df.ffill(limit=2).mask(df.bfill().isna())
  2. out = out.mask(out.ne(df) & out.notna(),0)
  3. Out[83]: 
  4.     col1   col2
  5. 0  100.0    NaN
  6. 1    0.0  100.0
  7. 2    0.0    0.0
  8. 3  100.0    0.0
  9. 4    NaN    NaN
  10. 5    NaN  100.0
  11. 6    NaN    NaN

huangapple
  • 本文由 发表于 2023年6月29日 22:35:38
  • 转载请务必保留本文链接：https://go.coder-hub.com/76582088.html
  • dataframe
  • fillna
  • interpolation
  • pandas
  • python
匿名

发表评论

匿名网友

:?: :razz: :sad: :evil: :!: :smile: :oops: :grin: :eek: :shock: :???: :cool: :lol: :mad: :twisted: :roll: :wink: :idea: :arrow: :neutral: :cry: :mrgreen:

确定