tacotron2

逻辑和

tf.reduce_all(input_tensor ,  
    axis = None ,  # 要减少的维度
    keep_dims = False ,  
    name = None ,  
    reduction_indices = None
  )

逻辑或

tf.reduce_any(input_tensor ,  
    axis = None ,  # 要减少的维度
    keep_dims = False ,  
    name = None ,  
    reduction_indices = None)

tf.mod(对应元素相除，取余数）
tf.equal(对比矩阵相应的元素，相同True，否则false)

tf.cond(tf.equal(remainder, tf.zeros(tf.shape(remainder), dtype=tf.int32)),
                lambda: x,
                lambda: x + multiple - remainder)

(控制数据流向，tf.equal=true—x; else— x+multiple-remainder

tf.sequence_mask(lengths,
    maxlen=None,
    dtype=tf.bool,
    name=None)

举例子说明

import tensorflow as tf
 
a = tf.sequence_mask([1, 2, 3], 5)
b = tf.sequence_mask([[1, 2], [3, 4]])
 
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print(sess.run(a))
    print(sess.run(b))

输出

[[ True False False False False]
 [ True  True False False False]
 [ True  True  True False False]]

解析：maxlen是5，所以一共有5列，lengths有三个元素[1,2,3]，所以有三行，每一行分别前1、2、3个元素为True

[[[ True False False False]
  [ True  True False False]]

 [[ True  True  True False]
  [ True  True  True  True]]]

解析：因为没有指定maxlen，故maxlen默认取lengths中的最大值4，所以一共有4列，lengths是二维数组，将其看作由两个一维lengths组成，所以输出也可以看作由这两个一维lengths的输出所组成

查看tensor的维度

x=tf.constant([0,1,2])
x_shape = tf.shape(x)
print(sess.run(x_shape))

loss = tf.losses.mean_squared_error(labels=mel_targets, 
                             predictions=outputs, 
                             weights=weights)

实际进行的操作是 label-prediction,然后按照weights归一求均方差
举个例子

mel_targets = tf.constant([[0,1,2,3],[4,5,6,7],[8,9,10,11]])
outputs = tf.constant([[1,1,1,1],[1,1,1,1],[1,1,1,1]])
weights = tf.constant([[1,1,1,1],[1,1,1,1],[0,0,0,0]])
res = tf.losses.mean_squared_error(labels=mel_targets, predictions=outputs, weights=weights)
sess = tf.Session()
print(sess.run(res))

输出11.5，实际上是对

label-output,然后前边行不变，最后一行为0，再求原来非零元素的均方差