在Tensorflow中将一组常数(一维数组)与一组矩阵(三维数组)相乘

时间:2018-08-29 23:04:29

标签: python tensorflow matrix matrix-multiplication

我的最终目标是训练以均值和协方差为参数的4D多元高斯分布

enter image description here

哪里

enter image description here

enter image description here

当前我有以下代码:

import tensorflow as tf
import numpy as np

value = [[1,2.0,3,4,5],[0,2,4,6,8],[80,7,6,5,4]]

value=tf.constant(value)

cov= tf.slice(value,[0,int(value.shape[1])-1],[int(value.shape[0]),1])
mean= tf.slice(value,[0,0],[int(value.shape[0]),int(value.shape[1])-1])

eyes=tf.eye(int(mean.shape[1]),batch_shape=[int(value.shape[0])])


#eyes = tf.multiply(eyes,cov)



normal = tf.contrib.distributions.MultivariateNormalFullCovariance(
                 loc=mean,
                 covariance_matrix=eyes) 

value = [[1,2.0,3,4,5],[0,2,4,6,8],[80,7,6,5,4]]是其余代码可能正在接收的示例。

在上面的示例中

cov = <tf.Tensor 'Slice_2:0' shape=(3, 1) dtype=float32>
eyes = <tf.Tensor 'eye_1/MatrixDiag:0' shape=(3, 4, 4) dtype=float32>

                     cov =  [[5.] [8.] [4.]]` 
                     eyes =  [[[1. 0. 0. 0.]
                              [0. 1. 0. 0.]
                              [0. 0. 1. 0.]
                              [0. 0. 0. 1.]]

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

                             [[1. 0. 0. 0.]
                              [0. 1. 0. 0.]
                              [0. 0. 1. 0.]
                              [0. 0. 0. 1.]]]`

我的问题是,在给定resultcov的情况下,如何获得eyes?结果如下:

result = [[[5., 0., 0., 0.],
           [0., 5., 0., 0.],
           [0., 0., 5., 0.],
           [0., 0., 0., 5.]],

          [[8., 0., 0., 0.],
           [0., 8., 0., 0.],
           [0., 0., 8., 0.],
           [0., 0., 0., 8.]],

          [[4., 0., 0., 0.],
           [0., 4., 0., 0.],
           [0., 0., 4., 0.],
           [0., 0., 0., 4.]]]

enter image description here

预先感谢

1 个答案:

答案 0 :(得分:1)

Tensorflow使用与numpy相同的索引类型,这可能非常强大。

您可以在此处查看详细信息:https://docs.scipy.org/doc/numpy-1.13.0/user/basics.broadcasting.html请注意,np.newaxis的定义与None相同。

对于您的问题,您可以在数据中添加额外的维度,以确保数组的乘法运算没有歧义。

import numpy as np
cov = np.array([[5.],[8.],[4.]])
eyes = np.array([[[1.0,0,0,0],[0.0,1.0,0.0,0],[0.0,0.0,1.0,0],[0.0,0.0,0.0,1.0]],[[1.0,0,0,0],[0.0,1.0,0.0,0],[0.0,0.0,1.0,0],[0.0,0.0,0.0,1.0]],[[1.0,0,0,0],[0.0,1.0,0.0,0],[0.0,0.0,1.0,0],[0.0,0.0,0.0,1.0]]])
result = cov[:,:,None]*eyes

在此处使用None会增加一个额外的维度,使cov成为3x1x1数组,可以明确地与3x4x4数组相乘。您也可以在tensorflow中以这种方式使用None

如果每个对应维度的大小相同或其中一个的大小为1,则两个数组可以明确相乘。