一个在numpy中使用矢量化并在pandas中应用的简单示例:

import numpy as np

a = np.array(['male','female','female','male'])

#define function
onehot_function = lambda x: 1.0 if (x=='male') else 0.0

onehot_a = np.vectorize(onehot_function)(a)

print(onehot_a)
# [1., 0., 0., 1.]

# -----------------------------------------

import pandas as pd

s = pd.Series(['male','female','female','male'])
onehot_s = s.apply(onehot_function)

print(onehot_s)
# 0    1.0
# 1    0.0
# 2    0.0
# 3    1.0
# dtype: float64

我知道我来晚了，但是以自动方式热编码数据框架的最简单方法是使用这个函数:

def hot_encode(df):
    obj_df = df.select_dtypes(include=['object'])
    return pd.get_dummies(df, columns=obj_df.columns).values

首先，最简单的热编码方法:使用Sklearn。

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

其次，我不认为使用熊猫进行一个热编码是那么简单(虽然未经证实)

在pandas中为python创建虚拟变量

最后，你需要一个热编码吗?一个热编码以指数方式增加了特征的数量，大大增加了任何分类器或任何你要运行的东西的运行时间。特别是当每个分类特征都有很多层次时。相反，你可以进行虚拟编码。

使用虚拟编码通常工作得很好，运行时间和复杂性要少得多。一位睿智的教授曾经告诉我，“少即是多”。

如果你愿意，这是我的自定义编码函数的代码。

from sklearn.preprocessing import LabelEncoder

#Auto encodes any dataframe column of type category or object.
def dummyEncode(df):
        columnsToEncode = list(df.select_dtypes(include=['category','object']))
        le = LabelEncoder()
        for feature in columnsToEncode:
            try:
                df[feature] = le.fit_transform(df[feature])
            except:
                print('Error encoding '+feature)
        return df

编辑:比较更清楚:

一热编码:将n层转换为n-1列。

Index  Animal         Index  cat  mouse
  1     dog             1     0     0
  2     cat       -->   2     1     0
  3    mouse            3     0     1

你可以看到，如果你的分类特征中有许多不同类型(或级别)，这会使你的记忆爆发式增长。记住，这只是一列。

伪代码:

Index  Animal         Index  Animal
  1     dog             1      0   
  2     cat       -->   2      1 
  3    mouse            3      2

转换为数字表示。极大地节省了特征空间，代价是准确性。

我在我的声学模型中使用了这个: 也许这对你的模型有帮助。

def one_hot_encoding(x, n_out):
    x = x.astype(int)  
    shape = x.shape
    x = x.flatten()
    N = len(x)
    x_categ = np.zeros((N,n_out))
    x_categ[np.arange(N), x] = 1
    return x_categ.reshape((shape)+(n_out,))

扩展@Martin Thoma的答案

def one_hot_encode(y):
    """Convert an iterable of indices to one-hot encoded labels."""
    y = y.flatten() # Sometimes not flattened vector is passed e.g (118,1) in these cases
    # the function ends up creating a tensor e.g. (118, 2, 1). flatten removes this issue
    nb_classes = len(np.unique(y)) # get the number of unique classes
    standardised_labels = dict(zip(np.unique(y), np.arange(nb_classes))) # get the class labels as a dictionary
    # which then is standardised. E.g imagine class labels are (4,7,9) if a vector of y containing 4,7 and 9 is
    # directly passed then np.eye(nb_classes)[4] or 7,9 throws an out of index error.
    # standardised labels fixes this issue by returning a dictionary;
    # standardised_labels = {4:0, 7:1, 9:2}. The values of the dictionary are mapped to keys in y array.
    # standardised_labels also removes the error that is raised if the labels are floats. E.g. 1.0; element
    # cannot be called by an integer index e.g y[1.0] - throws an index error.
    targets = np.vectorize(standardised_labels.get)(y) # map the dictionary values to array.
    return np.eye(nb_classes)[targets]

在这里我尝试了这个方法:

import numpy as np
#converting to one_hot





def one_hot_encoder(value, datal):

    datal[value] = 1

    return datal


def _one_hot_values(labels_data):
    encoded = [0] * len(labels_data)

    for j, i in enumerate(labels_data):
        max_value = [0] * (np.max(labels_data) + 1)

        encoded[j] = one_hot_encoder(i, max_value)

    return np.array(encoded)