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,))

你可以用numpy来做。眼和一个使用数组元素的选择机制:

import numpy as np
nb_classes = 6
data = [[2, 3, 4, 0]]

def indices_to_one_hot(data, nb_classes):
    """Convert an iterable of indices to one-hot encoded labels."""
    targets = np.array(data).reshape(-1)
    return np.eye(nb_classes)[targets]

indices_to_one_hot(nb_classes, data)的返回值现在是

array([[[ 0.,  0.,  1.,  0.,  0.,  0.],
        [ 0.,  0.,  0.,  1.,  0.,  0.],
        [ 0.,  0.,  0.,  0.,  1.,  0.],
        [ 1.,  0.,  0.,  0.,  0.,  0.]]])

. remodeling(-1)的作用是确保标签格式正确(也可能有[[2]，[3]，[4]，[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]

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。

假设在10个变量中，在数据帧中有3个分类变量，分别为cname1、cname2和cname3。 然后下面的代码将自动在新的数据框架中创建一个热编码变量。

import category_encoders as ce
encoder_var=ce.OneHotEncoder(cols=['cname1','cname2','cname3'],handle_unknown='return_nan',return_df=True,use_cat_names=True)
new_df = encoder_var.fit_transform(old_df)

为了补充其他问题，让我提供如何使用Numpy使用Python 2.0函数:

def one_hot(y_):
    # Function to encode output labels from number indexes 
    # e.g.: [[5], [0], [3]] --> [[0, 0, 0, 0, 0, 1], [1, 0, 0, 0, 0, 0], [0, 0, 0, 1, 0, 0]]

    y_ = y_.reshape(len(y_))
    n_values = np.max(y_) + 1
    return np.eye(n_values)[np.array(y_, dtype=np.int32)]  # Returns FLOATS

行n_values = np.max(y_) + 1可以硬编码，以便在使用小批量的情况下使用足够数量的神经元。

使用此函数的演示项目/教程: https://github.com/guillaume-chevalier/LSTM-Human-Activity-Recognition