TensorFlow 全连接层 tf.layers.dense()

函数原型：

tf.layers.dense(
	inputs,
	units,
	activation=None,
	use_bias=True,
	kernel_initializer=None,
	bias_initializer=tf.zeros_initializer(),
	kernel_regularizer=None,
	bias_regularizer=None,
	activity_regularizer=None,
	kernel_constraint=None,
	bias_constraint=None,
	trainable=True,
	name=None,
	reuse=None 
	)

---

举例：

假设有维度为4 的输入向量 (3, 4, 5, 6)，要经过全连接层计算得到输出为2的向量， 对应的权重为(1, 1, 1, 1) 和 (2, 2, 2, 2), 全连接层的计算示意图如下：

矩阵表示为：

$$output=input\times W^T=\left[\begin{array}{cccc}3& 4& 5& 6\end{array}\right]\left[\begin{array}{cc}1& 2\\ 1& 2\\ 1& 2\\ 1& 2\end{array}\right]=\left[\begin{array}{cc}18& 36\end{array}\right]$$

import tensorflow as tf
import numpy as np

# input_shape = [C, W, H]
input = [[1, 2, 3, 4]]

# input_shape = [input_channel;, width, height]
weights = [[1, 1, 1, 1],
           [2, 2, 2, 2]]

input_tf = tf.constant(input, dtype=tf.float32)
weights = np.transpose(weights, (1, 0))
weights_init = tf.constant_initializer(weights)

output_tf = tf.layers.dense(input_tf, 2, kernel_initializer=weights_init)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    output = sess.run(output_tf)
    print(output)


# 输出为：
# [[10. 20.]]