pandas进行one-hot编码

数值大小无意义

如颜色，使用内置函数hash()

import pandas as pd


def onehot_hash(df, column):
    '''进行数值大小无意义的独热编码

    :type df: pandas.core.frame.DataFrame
    :type column: str
    '''
    D = 2 ** 24
    df[column] = df[column].apply(lambda x: abs(hash(str(x)) % D))


if __name__ == '__main__':
    df = pd.DataFrame([['red'], ['green'], ['blue'], ['red']])
    df.columns = ['color']
    onehot_hash(df, 'color')
    print(df)
#       color
# 0    704795
# 1   9649319
# 2  16714041
# 3    704795

数值大小有意义

如尺码，使用映射pandas.core.series.Series.map(arg, na_action=None)

import pandas as pd


def onehot_map(df, column, mapping):
    '''进行数值大小有意义的独热编码

    :type df: pandas.core.frame.DataFrame
    :type column: str
    :type mapping: dict
    '''
    df[column] = df[column].map(mapping)


if __name__ == '__main__':
    df = pd.DataFrame([['M'], ['L'], ['XL']])
    df.columns = ['size']
    onehot_map(df, 'size', {'M': 1, 'L': 2, 'XL': 3})
    print(df)
#    size
# 0     1
# 1     2
# 2     3

get_dummies

将分类变量转换为哑变量

import pandas as pd

df = pd.DataFrame([['red'], ['blue'], ['green']])
df.columns = ['color']
print(pd.get_dummies(df))
#    color_blue  color_green  color_red
# 0           0            0          1
# 1           1            0          0
# 2           0            1          0

参考文献

1. pandas使用get_dummies进行one-hot编码
2. FTRL starter code | Kaggle