我有一个从数据透视表创建的数据框,看起来类似于:
import pandas as pd
d = {('company1', 'False Negative'): {'April- 2012': 112.0, 'April- 2013': 370.0, 'April- 2014': 499.0, 'August- 2012': 431.0, 'August- 2013': 496.0, 'August- 2014': 221.0},
('company1', 'False Positive'): {'April- 2012': 0.0, 'April- 2013': 544.0, 'April- 2014': 50.0, 'August- 2012': 0.0, 'August- 2013': 0.0, 'August- 2014': 426.0},
('company1', 'True Positive'): {'April- 2012': 0.0, 'April- 2013': 140.0, 'April- 2014': 24.0, 'August- 2012': 0.0, 'August- 2013': 0.0,'August- 2014': 77.0},
('company2', 'False Negative'): {'April- 2012': 112.0, 'April- 2013': 370.0, 'April- 2014': 499.0, 'August- 2012': 431.0, 'August- 2013': 496.0, 'August- 2014': 221.0},
('company2', 'False Positive'): {'April- 2012': 0.0, 'April- 2013': 544.0, 'April- 2014': 50.0, 'August- 2012': 0.0, 'August- 2013': 0.0, 'August- 2014': 426.0},
('company2', 'True Positive'): {'April- 2012': 0.0, 'April- 2013': 140.0, 'April- 2014': 24.0, 'August- 2012': 0.0, 'August- 2013': 0.0,'August- 2014': 77.0},}
df = pd.DataFrame(d)
company1 company2
FN FP TP FN FP TP
April- 2012 112 0 0 112 0 0
April- 2013 370 544 140 370 544 140
April- 2014 499 50 24 499 50 24
August- 2012 431 0 0 431 0 0
August- 2013 496 0 0 496 0 0
August- 2014 221 426 77 221 426 77
我希望在multiindex列的上层迭代,将每个公司除以它的总和来创建一个百分比:
company1 company2
FN FP TP FN FP TP
April- 2012 1 0 0 1 0 0
April- 2013 .35 .51 .13 .35 .51 .13
April- 2014 .87 .09 .03 .87 .09 .03
etc.
我事先并不知道公司名称。这是昨天提出的问题的变体:Summing multiple columns with multiindex columns
答案 0 :(得分:4)
您可以使用div
方法除以总和(您可以指定要匹配的级别):
df.div(df.sum(axis=1, level=0), level=0)
答案 1 :(得分:1)
虽然joris的解决方案运行良好,但我想补充说,只要Multiindex具有更多级别,这将无法正常工作。以下是我的解决方案,部分基于StackOverflow帖子(Normalize DataFrame by group)和Pandas文档(http://pandas.pydata.org/pandas-docs/stable/groupby.html)。
d = {
('X', 'company1', 'False Negative'): {'April- 2012': 112.0, 'April- 2013': 370.0, 'April- 2014': 499.0, 'August- 2012': 431.0, 'August- 2013': 496.0, 'August- 2014': 221.0},
('X', 'company1', 'False Positive'): {'April- 2012': 0.0, 'April- 2013': 544.0, 'April- 2014': 50.0, 'August- 2012': 0.0, 'August- 2013': 0.0, 'August- 2014': 426.0},
('X', 'company1', 'True Positive'): {'April- 2012': 0.0, 'April- 2013': 140.0, 'April- 2014': 24.0, 'August- 2012': 0.0, 'August- 2013': 0.0,'August- 2014': 77.0},
('X', 'company2', 'False Negative'): {'April- 2012': 112.0, 'April- 2013': 370.0, 'April- 2014': 499.0, 'August- 2012': 431.0, 'August- 2013': 496.0, 'August- 2014': 221.0},
('X', 'company2', 'False Positive'): {'April- 2012': 0.0, 'April- 2013': 544.0, 'April- 2014': 50.0, 'August- 2012': 0.0, 'August- 2013': 0.0, 'August- 2014': 426.0},
('X', 'company2', 'True Positive'): {'April- 2012': 0.0, 'April- 2013': 140.0, 'April- 2014': 24.0, 'August- 2012': 0.0, 'August- 2013': 0.0,'August- 2014': 77.0},
('Y','company1', 'False Negative'): {'April- 2012': 112.0, 'April- 2013': 370.0, 'April- 2014': 499.0, 'August- 2012': 431.0, 'August- 2013': 496.0, 'August- 2014': 221.0},
('Y','company1', 'False Positive'): {'April- 2012': 0.0, 'April- 2013': 544.0, 'April- 2014': 50.0, 'August- 2012': 0.0, 'August- 2013': 0.0, 'August- 2014': 426.0},
('Y','company1', 'True Positive'): {'April- 2012': 0.0, 'April- 2013': 140.0, 'April- 2014': 24.0, 'August- 2012': 0.0, 'August- 2013': 0.0,'August- 2014': 77.0},
('Y','company2', 'False Negative'): {'April- 2012': 112.0, 'April- 2013': 370.0, 'April- 2014': 499.0, 'August- 2012': 431.0, 'August- 2013': 496.0, 'August- 2014': 221.0},
('Y','company2', 'False Positive'): {'April- 2012': 0.0, 'April- 2013': 544.0, 'April- 2014': 50.0, 'August- 2012': 0.0, 'August- 2013': 0.0, 'August- 2014': 426.0},
('Y','company2', 'True Positive'): {'April- 2012': 0.0, 'April- 2013': 140.0, 'April- 2014': 24.0, 'August- 2012': 0.0, 'August- 2013': 0.0,'August- 2014': 77.0},
}
df = pd.DataFrame(d)
# extrapolation of original method: not working!
# df.div(df.sum(axis=1,level=[0,1]), level=[0,1]) # does not work
# alternative 1: replicating the sums for each company to fit the number of columns using numpy
df.div(np.repeat(df.sum(axis=1,level=[0,1]).values, 3, axis=1), axis=1)
# alternative 2: stacking, grouping, transforming and unstacking
df.columns.names = ['top', 'company', 'result'] # naming column levels for convenience
df.\
stack(["top", "company", "result"]).\
groupby(level=[0,1,2]).\
transform(lambda x: (x / x.sum(axis=0))).\
unstack(["top", "company", "result"])