在实际建模过程中，我们经常需要对离散型变量进行encode处理，譬如性别，类目，标签等等。通过one hot encode方法进行处理。经验表明，这些操作处理对提升预测模型性能有帮助。

http://en.wikipedia.org/wiki/One-hot

# -*- coding: utf-8 -*-

""" Small script that shows hot to do one hot encoding

    of categorical columns in a pandas DataFrame.

    See:

    http://scikit-learn.org/dev/modules/generated/sklearn.preprocessing.OneHotEncoder.html#sklearn.preprocessing.OneHotEncoder

    http://scikit-learn.org/dev/modules/generated/sklearn.feature_extraction.DictVectorizer.html

"""

import pandas

import random

import numpy

from sklearn.feature_extraction import DictVectorizer

def one_hot_dataframe(data, cols, replace=False):

    """ Takes a dataframe and a list of columns that need to be encoded.

        Returns a 3-tuple comprising the data, the vectorized data,

        and the fitted vectorizor.

    """

    vec = DictVectorizer()

    mkdict = lambda row: dict((col, row[col]) for col in cols)

    # manuplate the column

    vecData = pandas.DataFrame(vec.fit_transform(data[cols].apply(mkdict, axis=1)).toarray())

    # get column names

    vecData.columns = vec.get_feature_names()

    vecData.index = data.index

    if replace is True:

        data = data.drop(cols, axis=1)

        #column join based on index

        data = data.join(vecData)

    return (data, vecData, vec)

def main():

    # Get a random DataFrame

    df = pandas.DataFrame(numpy.random.randn(25, 3), columns=['a', 'b', 'c'])

    # Make some random categorical columns

    df['e'] = [random.choice(('Chicago', 'Boston', 'New York')) for i in range(df.shape[0])]

    df['f'] = [random.choice(('Chrome', 'Firefox', 'Opera', "Safari")) for i in range(df.shape[0])]

    print df

    # Vectorize the categorical columns: e & f

    df, _, _ = one_hot_dataframe(df, ['e', 'f'], replace=True)

    print df

if __name__ == '__main__':

    main()

原始数据

http://s8/mw690/001N0mEhgy6I30UKESza7&690
one-hot encode数据
http://s3/mw690/001N0mEhgy6I30UNNUS42&690