python - Python - 转换数据框和切片

Question

我附上了一个截图来帮助解释。我有一个从克利夫兰心脏数据集中提取的数据框，它需要 76 列并将它们放入 7 列，并将附加列包装到下一行。我试图弄清楚如何将该数据框转换为可读格式，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会有所不同。我以为我可以使用 data.loc[:, :'xyz'] 开始，但我不确定从这里去哪里：

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

然后，我将不得不从那里为这些变量分配列名。令人惊讶的是，一旦我解决了这个问题，训练、测试、验证部分就会变得容易得多。在此先感谢您的帮助。（我是菜鸟）

标签： pythonpandas

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