python-如何使用熊猫融化的值和它的错误

我有以下广泛的数据集：

import pandas as pd
from io import StringIO
testcsv = """P,N,N_relerr,F,F_relerr
10,6073.98,0.0022,61.973,0.0036
12,6412.97,0.0021,65.405,0.0036
4,4141.24,0.0019,42.8202,0.0032
6,5009.83,51.9615,0.0031
8,5601.87,0.0025,57.8129,0.0042"""
csvfile = StringIO(testcsv)
df = pd.read_csv(csvfile)

    P   N           N_relerr  F         F_relerr
0   10  6073.98     0.0022    61.9730   0.0036
1   12  6412.97     0.0021    65.4050   0.0036
2   4   4141.24     0.0019    42.8202   0.0032
3   6   5009.83     0.0019    51.9615   0.0031
4   8   5601.87     0.0025    57.8129   0.0042

我想变成一个长数据集,该数据集具有“计数”(N和F列)以及相关的错误(N_relerr和F_relerr)：

    P   which   count       err
0   10  N       6073.9800   0.0022
1   12  N       6412.9700   0.0021
2   4   N       4141.2400   0.0019
3   6   N       5009.8300   0.0019
4   8   N       5601.8700   0.0025
5   10  F       61.9730     0.0036
6   12  F       65.4050     0.0036
7   4   F       42.8202     0.0032
8   6   F       51.9615     0.0031
9   8   F       57.8129     0.0042

因为这是格式,所以我需要使用带有’N’和’F’计数彼此区分的plotnine绘制误差线.我当前非常难看的解决方案是：

dflong = (df[['P','N','F']]
           .melt(id_vars=['P'],var_name='which',value_name='count'))
dferr = (df[['P','N_relerr','F_relerr']]
          .melt(id_vars=['P'],value_name='count_relerr'))
dflong['err'] = dferr['count_relerr'].copy()

我的猜测是,有一个优雅的方法可以使用multiindex列以及堆栈,从看起来像这样的数据集开始：

            N                   F
    P       counts    relerr    counts    relerr
0   10      6073.98   0.0022    61.9730   0.0036
1   12      6412.97   0.0021    65.4050   0.0036
2   4       4141.24   0.0019    42.8202   0.0032
3   6       5009.83   0.0019    51.9615   0.0031
4   8       5601.87   0.0025    57.8129   0.0042

我可以通过以下方式创建该数据框：

cols = {'P': 'P','N': ('N','counts'),'N_relerr': ('N',"relerr"),'F': ('F','F_relerr': ('F','relerr')}
nested_df = df.rename(columns=cols)
nested_df.columns = [c if isinstance(c,tuple) 
                     else ('',c) for c in nested_df.columns]
nested_df.columns = pd.MultiIndex.from_tuples(nested_df.columns) 

(我认为必须有一个更好的方法),但是我还没有弄清楚如何有效地使用堆栈来获得我想要的东西.

有人知道规范的解决方案吗？谢谢！

最佳答案

您可以使用pd.wide_to_long,非常适合那种“同时融化”的情况,只需稍微重命名列即可.

import pandas as pd
from io import StringIO
testcsv = """P,0.0042"""
csvfile = StringIO(testcsv)
df = pd.read_csv(csvfile)

#Rename columns with set_axis
d1 = df.set_axis(['P','Count_N','Err_N','Count_F','Err_F'],axis=1,inplace=False)

#Use pd.wide_to_long to reshape dataframe
pd.wide_to_long(d1,['Count','Err'],'P','which',sep='_',suffix='.+')

输出：

              Count     Err
P  which                   
10 N      6073.9800  0.0022
12 N      6412.9700  0.0021
4  N      4141.2400  0.0019
6  N      5009.8300  0.0019
8  N      5601.8700  0.0025
10 F        61.9730  0.0036
12 F        65.4050  0.0036
4  F        42.8202  0.0032
6  F        51.9615  0.0031
8  F        57.8129  0.0042