试试这个:

!pip install category_encoders
import category_encoders as ce

categorical_columns = [...the list of names of the columns you want to one-hot-encode ...]
encoder = ce.OneHotEncoder(cols=categorical_columns, use_cat_names=True)
df_train_encoded = encoder.fit_transform(df_train_small)

df_encoded.head ()

生成的数据框架df_train_encoded与原始数据框架相同，但是分类特征现在被它们的单热编码版本所取代。

更多关于category_encoders的信息请点击这里。

您可以将数据传递给catboost分类器而不进行编码。Catboost通过执行单热和目标扩展平均编码来处理分类变量本身。

熊猫的热编码非常简单:

def one_hot(df, cols):
    """
    @param df pandas DataFrame
    @param cols a list of columns to encode 
    @return a DataFrame with one-hot encoding
    """
    for each in cols:
        dummies = pd.get_dummies(df[each], prefix=each, drop_first=False)
        df = pd.concat([df, dummies], axis=1)
    return df

编辑:

使用sklearn的LabelBinarizer实现one_hot的另一种方法:

from sklearn.preprocessing import LabelBinarizer 
label_binarizer = LabelBinarizer()
label_binarizer.fit(all_your_labels_list) # need to be global or remembered to use it later

def one_hot_encode(x):
    """
    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 label_binarizer.transform(x)

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。

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]