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)

这对我来说很管用:

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

输出:

[0, 1, 2, 0]

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。

它可以而且应该很简单:

class OneHotEncoder:
    def __init__(self,optionKeys):
        length=len(optionKeys)
        self.__dict__={optionKeys[j]:[0 if i!=j else 1 for i in range(length)] for j in range(length)}

用法:

ohe=OneHotEncoder(["A","B","C","D"])
print(ohe.A)
print(ohe.D)

使用Pandas进行基本的单热编码要容易得多。如果您正在寻找更多的选项，您可以使用scikit-learn。

对于Pandas的基本单热编码，您可以将数据帧传递给get_dummies函数。

例如，如果我有一个名为imdb_movies的数据帧:

.．.和我想要一个热编码的评级列，我这样做:

pd.get_dummies(imdb_movies.Rated)

这将返回一个新的数据框架，其中包含一个列，表示存在的每个评级“级别”，以及一个1或0，指定给定观察值的评级。

通常，我们希望它是原始数据框架的一部分。在本例中，我们使用“列绑定”将新的虚拟编码框架附加到原始框架上。

我们可以使用Pandas concat函数进行列绑定:

rated_dummies = pd.get_dummies(imdb_movies.Rated)
pd.concat([imdb_movies, rated_dummies], axis=1)

现在我们可以对完整的数据框架进行分析。

简单效用函数

我建议你自己做一个效用函数来快速做到这一点:

def encode_and_bind(original_dataframe, feature_to_encode):
    dummies = pd.get_dummies(original_dataframe[[feature_to_encode]])
    res = pd.concat([original_dataframe, dummies], axis=1)
    return(res)

用法:

encode_and_bind(imdb_movies, 'Rated')

结果:

另外，根据@pmalbu的评论，如果你想让函数删除原来的feature_to_encode，那么使用这个版本:

def encode_and_bind(original_dataframe, feature_to_encode):
    dummies = pd.get_dummies(original_dataframe[[feature_to_encode]])
    res = pd.concat([original_dataframe, dummies], axis=1)
    res = res.drop([feature_to_encode], axis=1)
    return(res) 

你可以在同一时间编码多个特征，如下所示:

features_to_encode = ['feature_1', 'feature_2', 'feature_3',
                      'feature_4']
for feature in features_to_encode:
    res = encode_and_bind(train_set, feature)

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]