只需在TensorFlow中查找等效的np.std()来计算张量的标准偏差。
答案 0 :(得分:39)
答案 1 :(得分:6)
您还可以在以下由Keras改编的代码中使用reduce_std
:
#coding=utf-8
import numpy as np
import tensorflow as tf
def reduce_var(x, axis=None, keepdims=False):
"""Variance of a tensor, alongside the specified axis.
# Arguments
x: A tensor or variable.
axis: An integer, the axis to compute the variance.
keepdims: A boolean, whether to keep the dimensions or not.
If `keepdims` is `False`, the rank of the tensor is reduced
by 1. If `keepdims` is `True`,
the reduced dimension is retained with length 1.
# Returns
A tensor with the variance of elements of `x`.
"""
m = tf.reduce_mean(x, axis=axis, keep_dims=True)
devs_squared = tf.square(x - m)
return tf.reduce_mean(devs_squared, axis=axis, keep_dims=keepdims)
def reduce_std(x, axis=None, keepdims=False):
"""Standard deviation of a tensor, alongside the specified axis.
# Arguments
x: A tensor or variable.
axis: An integer, the axis to compute the standard deviation.
keepdims: A boolean, whether to keep the dimensions or not.
If `keepdims` is `False`, the rank of the tensor is reduced
by 1. If `keepdims` is `True`,
the reduced dimension is retained with length 1.
# Returns
A tensor with the standard deviation of elements of `x`.
"""
return tf.sqrt(reduce_var(x, axis=axis, keepdims=keepdims))
if __name__ == '__main__':
x_np = np.arange(10).reshape(2, 5).astype(np.float32)
x_tf = tf.constant(x_np)
with tf.Session() as sess:
print(sess.run(reduce_std(x_tf, keepdims=True)))
print(sess.run(reduce_std(x_tf, axis=0, keepdims=True)))
print(sess.run(reduce_std(x_tf, axis=1, keepdims=True)))
print(np.std(x_np, keepdims=True))
print(np.std(x_np, axis=0, keepdims=True))
print(np.std(x_np, axis=1, keepdims=True))