简短的回答

这里有一个函数，可以在不使用numpy、pandas或其他包的情况下进行一次性编码。它接受一个整数、布尔值或字符串(也可能是其他类型)的列表。

import typing


def one_hot_encode(items: list) -> typing.List[list]:
    results = []
    # find the unique items (we want to unique items b/c duplicate items will have the same encoding)
    unique_items = list(set(items))
    # sort the unique items
    sorted_items = sorted(unique_items)
    # find how long the list of each item should be
    max_index = len(unique_items)

    for item in items:
        # create a list of zeros the appropriate length
        one_hot_encoded_result = [0 for i in range(0, max_index)]
        # find the index of the item
        one_hot_index = sorted_items.index(item)
        # change the zero at the index from the previous line to a one
        one_hot_encoded_result[one_hot_index] = 1
        # add the result
        results.append(one_hot_encoded_result)

    return results

例子:

one_hot_encode([2, 1, 1, 2, 5, 3])

# [[0, 1, 0, 0],
#  [1, 0, 0, 0],
#  [1, 0, 0, 0],
#  [0, 1, 0, 0],
#  [0, 0, 0, 1],
#  [0, 0, 1, 0]]

one_hot_encode([True, False, True])

# [[0, 1], [1, 0], [0, 1]]

one_hot_encode(['a', 'b', 'c', 'a', 'e'])

# [[1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [1, 0, 0, 0], [0, 0, 0, 1]]

长(er)回答

I know there are already a lot of answers to this question, but I noticed two things. First, most of the answers use packages like numpy and/or pandas. And this is a good thing. If you are writing production code, you should probably be using robust, fast algorithms like those provided in the numpy/pandas packages. But, for the sake of education, I think someone should provide an answer which has a transparent algorithm and not just an implementation of someone else's algorithm. Second, I noticed that many of the answers do not provide a robust implementation of one-hot encoding because they do not meet one of the requirements below. Below are some of the requirements (as I see them) for a useful, accurate, and robust one-hot encoding function:

单热编码函数必须:

处理各种类型的列表(例如，整数，字符串，浮点数等)作为输入 处理带有重复项的输入列表 返回与输入相对应(与输入顺序相同)的列表的列表 返回一个列表的列表，其中每个列表都尽可能短

我测试了这个问题的许多答案，其中大多数都未能满足上述要求之一。

这对我来说很管用:

pandas.factorize( ['B', 'C', 'D', 'B'] )[0]

输出:

[0, 1, 2, 0]

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]

单热编码需要比将值转换为指示器变量更多的内容。通常ML过程要求您多次将此编码应用于验证或测试数据集，并将您构建的模型应用于实时观察数据。您应该存储用于构造模型的映射(转换)。一个好的解决方案是使用DictVectorizer或LabelEncoder(后面跟着get_dummies)。下面是一个你可以使用的函数:

def oneHotEncode2(df, le_dict = {}):
    if not le_dict:
        columnsToEncode = list(df.select_dtypes(include=['category','object']))
        train = True;
    else:
        columnsToEncode = le_dict.keys()   
        train = False;

    for feature in columnsToEncode:
        if train:
            le_dict[feature] = LabelEncoder()
        try:
            if train:
                df[feature] = le_dict[feature].fit_transform(df[feature])
            else:
                df[feature] = le_dict[feature].transform(df[feature])

            df = pd.concat([df, 
                              pd.get_dummies(df[feature]).rename(columns=lambda x: feature + '_' + str(x))], axis=1)
            df = df.drop(feature, axis=1)
        except:
            print('Error encoding '+feature)
            #df[feature]  = df[feature].convert_objects(convert_numeric='force')
            df[feature]  = df[feature].apply(pd.to_numeric, errors='coerce')
    return (df, le_dict)

这适用于pandas数据框架，它为数据框架的每一列创建并返回一个映射。所以你可以这样称呼它:

train_data, le_dict = oneHotEncode2(train_data)

然后在测试数据上，通过传递训练返回的字典进行调用:

test_data, _ = oneHotEncode2(test_data, le_dict)

一个等效的方法是使用DictVectorizer。我的博客上有一篇相关的文章。我在这里提到它是因为它为这种方法提供了一些理由，而不是简单地使用get_dummies post(披露:这是我自己的博客)。

您可以使用numpy。眼睛的功能。

import numpy as np

def one_hot_encode(x, n_classes):
    """
    One hot encode a list of sample labels. Return a one-hot encoded vector for each label.
    : x: List of sample Labels
    : return: Numpy array of one-hot encoded labels
     """
    return np.eye(n_classes)[x]

def main():
    list = [0,1,2,3,4,3,2,1,0]
    n_classes = 5
    one_hot_list = one_hot_encode(list, n_classes)
    print(one_hot_list)

if __name__ == "__main__":
    main()

结果

D:\Desktop>python test.py
[[ 1.  0.  0.  0.  0.]
 [ 0.  1.  0.  0.  0.]
 [ 0.  0.  1.  0.  0.]
 [ 0.  0.  0.  1.  0.]
 [ 0.  0.  0.  0.  1.]
 [ 0.  0.  0.  1.  0.]
 [ 0.  0.  1.  0.  0.]
 [ 0.  1.  0.  0.  0.]
 [ 1.  0.  0.  0.  0.]]

方法1:你可以使用pandas的pd.get_dummies。

示例1:

import pandas as pd
s = pd.Series(list('abca'))
pd.get_dummies(s)
Out[]: 
     a    b    c
0  1.0  0.0  0.0
1  0.0  1.0  0.0
2  0.0  0.0  1.0
3  1.0  0.0  0.0

示例2:

下面将把一个给定的列转换为一个hot。使用前缀有多个假人。

import pandas as pd
        
df = pd.DataFrame({
          'A':['a','b','a'],
          'B':['b','a','c']
        })
df
Out[]: 
   A  B
0  a  b
1  b  a
2  a  c

# Get one hot encoding of columns B
one_hot = pd.get_dummies(df['B'])
# Drop column B as it is now encoded
df = df.drop('B',axis = 1)
# Join the encoded df
df = df.join(one_hot)
df  
Out[]: 
       A  a  b  c
    0  a  0  1  0
    1  b  1  0  0
    2  a  0  0  1

方法2:使用Scikit-learn

使用OneHotEncoder的优点是能够拟合一些训练数据，然后使用相同的实例对一些其他数据进行转换。我们还有handle_unknown来进一步控制编码器对未见数据的处理。

给定一个具有三个特征和四个样本的数据集，我们让编码器找到每个特征的最大值，并将数据转换为二进制one-hot编码。

>>> from sklearn.preprocessing import OneHotEncoder
>>> enc = OneHotEncoder()
>>> enc.fit([[0, 0, 3], [1, 1, 0], [0, 2, 1], [1, 0, 2]])   
OneHotEncoder(categorical_features='all', dtype=<class 'numpy.float64'>,
   handle_unknown='error', n_values='auto', sparse=True)
>>> enc.n_values_
array([2, 3, 4])
>>> enc.feature_indices_
array([0, 2, 5, 9], dtype=int32)
>>> enc.transform([[0, 1, 1]]).toarray()
array([[ 1.,  0.,  0.,  1.,  0.,  0.,  1.,  0.,  0.]])

下面是这个例子的链接:http://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.OneHotEncoder.html