Python-转换数据框和切片

Question

我附上了截图以帮助解释。我从克利夫兰心脏数据集中提取了一个数据框，该数据框占用76列并将其放入7列，并将其他列包装到下一行。我试图弄清楚如何使该数据框成为可读的格式，如右侧数据框所示。

变量xyz将始终相同，但我列出的其他字母变量将不同。我以为我可以使用data.loc [:,：'xyz']来开始，但是我不确定从这里去哪里：

data = pd.read_csv("../resources/cleveland.data")
data.loc[:, :'xyz']

然后我将不得不从那里开始并将列名称分配给这些变量。出乎意料的是，一旦我弄清了这一点，培训，测试，验证部分将变得容易得多。先谢谢您的帮助。 （我是菜鸟）

Answer 1

输入数据

1   a   b   c
d   xyz 2   e
f   g   h   xyz
3   i   j   k

代码

import pandas as pd
import numpy as np

# The initial data doesn't contain header so set header to None
df = pd.read_csv("../resources/cleveland.data", header=None)
cols = df.columns.tolist()

# Reset the index to get the line number in the durty file
df = df.reset_index()

# After having melt the df, you can filter the df in order to have every values in one column.
# Those values are in the right order
df = pd.melt(df, id_vars=['index'], value_vars=cols)
df = df.sort_values(by=['index', 'variable'])

# Then you can set the line number
df['line'] = np.where(df.value == 'xyz', 1, np.nan)
df.line = df.line.cumsum()
df.line = df.line.bfill()

# If the file doesn't end with 'xyz', we have to set the line number to df.line.max() + 1
df.loc[df.line.isna(), 'line'] = df.line.max() + 1
df.line = df.line.ffill()

# We can set the column names as interger with a groupby cumsum
df['one'] = 1
df['col_name'] = df.groupby(['line'])['one'].cumsum()
df['col_name'] = "col_" + df['col_name'].astype('str')

# Then we can pivot the table
df = df[['value', 'line', 'col_name']]
df = df.pivot(index='line', columns='col_name', values='value')
print(df)

输出数据

col_name col_1 col_2 col_3 col_4 col_5 col_6
line
1.0          1     a     b     c     d   xyz
2.0          2     e     f     g     h   xyz
3.0          3     i     j     k   NaN   NaN

Answer 2

在将所有值组成一个大数组之后，使用numpy。 np.array_split + np.where的组合，用于在xyz之后分割索引：

样本数据：test.csv

1,a,b,c,d,e,f,g
h,i,j,k,xyz,2,a,b
c,d,e,f,g,h,i,j
k,xyz

代码

import numpy as np
import pandas as pd

arr = pd.read_csv('test.csv', header=None).values.ravel()

pd.DataFrame(np.array_split(arr, np.where(arr == 'xyz')[0]+1)).dropna(how='all')

输出：

  0  1  2  3  4  5  6  7  8  9  10 11   12
0  1  a  b  c  d  e  f  g  h  i  j  k  xyz
1  2  a  b  c  d  e  f  g  h  i  j  k  xyz

来自@CharlesR数据

   0  1  2  3     4     5
0  1  a  b  c     d   xyz
1  2  e  f  g     h   xyz
2  3  i  j  k  None  None