从零开始编写深度学习库（二）FullyconnecteLayer CPU编写

从零开始编写深度学习库（二）FullyconnecteLayer CPU编写
博客：http://blog.csdn.net/hjimce
微博：黄锦池-hjimce   qq:1393852684
一、C++实现
//y=x*w+b
static void CFullyconnecteLayer::forward(const Eigen::MatrixXf &inputs, const Eigen::MatrixXf &weights, const Eigen::VectorXf &bais
, Eigen::MatrixXf &outputs) {

outputs = inputs*weights;
outputs.rowwise() += bais.transpose();//每一行加上b
}
//y=x*w+b,反向求导后dw=x.T*dy,dx=dy*w.T,db=dy
static void CFullyconnecteLayer::backward(const Eigen::MatrixXf &inputs, const Eigen::MatrixXf &weights, const Eigen::VectorXf &bais,
const Eigen::MatrixXf &d_outputs,Eigen::MatrixXf &d_inputs, Eigen::MatrixXf &d_weights, Eigen::VectorXf &d_bais) {

d_weights = inputs.transpose()*d_outputs;
d_inputs = d_outputs*weights.transpose();
d_bais = d_outputs.colwise().sum();
}
static void CFullyconnecteLayer::test() {
int batch_size = 4;
int input_size = 3;
int output_size = 2;
Eigen::MatrixXf inputs(batch_size, input_size);
inputs <<1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12;
Eigen::MatrixXf weights(input_size, output_size);
weights << 0.55, 0.88, 0.75, 1.1, 0.11, 0.002;
Eigen::VectorXf bais(output_size);
bais << 3, 2;
Eigen::VectorXf label(batch_size);
label << 1, 0, 1, 1;

Eigen::MatrixXf outputs;//全连接层
forward(inputs, weights, bais, outputs);

Eigen::MatrixXf d_input, d_weights,d_output;
float loss;
CSoftmaxLayer::softmax_loss_forward_backward(outputs, label, d_output, loss);
std::cout << loss << std::endl;

Eigen::VectorXf d_bais;
backward(inputs, weights, bais, d_output, d_input, d_weights, d_bais);

std::cout << "outputs" << outputs << std::endl;
std::cout << "d_output" << d_output << std::endl;
std::cout << "d_input" << d_input << std::endl;
std::cout << "d_weights" << d_weights << std::endl;
std::cout << "d_bais"<<d_bais << std::endl;
}二、tensorflow验证：
import tensorflow as tf
inputs=tf.constant([[1,2,3],[4,5,6],[7,8,9],[10,11,12]],shape=(4,3),dtype=tf.float32)
weights=tf.constant([0.55, 0.88, 0.75, 1.1, 0.11, 0.002],shape=(3,2),dtype=tf.float32)
bais=tf.constant([3, 2],dtype=tf.float32)
label=tf.constant([1,0,1,1])

output=tf.matmul(inputs,weights)+bais
one_hot=tf.one_hot(label,2)
predicts=tf.nn.softmax(output)
loss =-tf.reduce_mean(one_hot * tf.log(predicts))

d_output,d_inputs,d_weights,d_bais=tf.gradients(loss,[output,inputs,weights,bais])

with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
loss_np,output_np,d_output_np, d_inputs_np, d_weights_np, d_bais_np=sess.run([loss,output,d_output,
d_inputs,d_weights,d_bais])
print (loss_np)
print ('output',output_np)
print('d_output', d_output_np)
print ("d_inputs",d_inputs_np)
print("d_weights", d_weights_np)
print("d_bais", d_bais_np)