keras中的keras.utils.to_categorical方法

参考链接：https://blog.csdn.net/nima1994/article/details/82468965

参考链接：https://blog.csdn.net/gdl3463315/article/details/82659378

to_categorical(y, num_classes=None, dtype='float32')

将整型的类别标签转为onehot编码。y为int数组，num_classes为标签类别总数，大于max(y)（标签从0开始的）。

返回：如果num_classes=None，返回len(y) * [max(y)+1]（维度，m*n表示m行n列矩阵，下同），否则为len(y) * num_classes。

1. import keras

2. ohl=keras.utils.to_categorical([1,3])

3. # ohl=keras.utils.to_categorical([[1],[3]])

4. print(ohl)

5. """

6. [[0. 1. 0. 0.]

7. [0. 0. 0. 1.]]

8. """

9. ohl=keras.utils.to_categorical([1,3],num_classes=5)

10. print(ohl)

11. """

12. [[0. 1. 0. 0. 0.]

13. [0. 0. 0. 1. 0.]]

14. """

该部分keras源码如下：

1. def to_categorical(y, num_classes=None, dtype='float32'):

2. """Converts a class vector (integers) to binary class matrix.

3. E.g. for use with categorical_crossentropy.

4. # Arguments

5. y: class vector to be converted into a matrix

6. (integers from 0 to num_classes).

7. num_classes: total number of classes.

8. dtype: The data type expected by the input, as a string

9. (`float32`, `float64`, `int32`...)

10. # Returns

11. A binary matrix representation of the input. The classes axis

12. is placed last.

13. """

14. y = np.array(y, dtype='int')

15. input_shape = y.shape

16. if input_shape and input_shape[-1] == 1 and len(input_shape) > 1:

17. input_shape = tuple(input_shape[:-1])

18. y = y.ravel()

19. if not num_classes:

20. num_classes = np.max(y) + 1

21. n = y.shape[0]

22. categorical = np.zeros((n, num_classes), dtype=dtype)

23. categorical[np.arange(n), y] = 1

24. output_shape = input_shape + (num_classes,)

25. categorical = np.reshape(categorical, output_shape)

26. return categorical

简单来说：**keras.utils.to_categorical函数：是把类别标签转换为onehot编码（categorical就是类别标签的意思，表示现实世界中你分类的各类别）， 而onehot编码是一种方便计算机处理的二元编码。**