Refactored connected to use blas

Makefile

@@ -1,5 +1,5 @@
 CC=gcc
-GPU=1
+GPU=0
 COMMON=-Wall -Wfatal-errors `pkg-config --cflags opencv` -I/usr/local/cuda/include/
 ifeq ($(GPU), 1) 
 COMMON+=-DGPU
@@ -7,6 +7,7 @@ else
 endif
 UNAME = $(shell uname)
 OPTS=-Ofast -flto
+OPTS=-Ofast -flto
 ifeq ($(UNAME), Darwin)
 COMMON+= -isystem /usr/local/Cellar/opencv/2.4.6.1/include/opencv -isystem /usr/local/Cellar/opencv/2.4.6.1/include
 ifeq ($(GPU), 1)

src/cnn.c

@@ -916,8 +916,8 @@ int main(int argc, char *argv[])
     //test_ensemble();
     //test_nist_single();
     //test_nist();
-    //train_nist();
+    train_nist();
-    test_convolutional_layer();
+    //test_convolutional_layer();
     //test_col2im();
     //test_cifar10();
     //train_cifar10();

src/connected_layer.c

@@ -26,7 +26,6 @@ connected_layer *make_connected_layer(int batch, int inputs, int outputs, ACTIVA
 
     layer->weight_updates = calloc(inputs*outputs, sizeof(float));
     //layer->weight_adapt = calloc(inputs*outputs, sizeof(float));
-    layer->weight_momentum = calloc(inputs*outputs, sizeof(float));
     layer->weights = calloc(inputs*outputs, sizeof(float));
     float scale = 1./inputs;
     scale = .05;
@@ -35,7 +34,6 @@ connected_layer *make_connected_layer(int batch, int inputs, int outputs, ACTIVA
 
     layer->bias_updates = calloc(outputs, sizeof(float));
     //layer->bias_adapt = calloc(outputs, sizeof(float));
-    layer->bias_momentum = calloc(outputs, sizeof(float));
     layer->biases = calloc(outputs, sizeof(float));
     for(i = 0; i < outputs; ++i){
         //layer->biases[i] = rand_normal()*scale + scale;
@@ -50,24 +48,19 @@ connected_layer *make_connected_layer(int batch, int inputs, int outputs, ACTIVA
 
 void update_connected_layer(connected_layer layer)
 {
-    int i;
+    axpy_cpu(layer.outputs, layer.learning_rate, layer.bias_updates, 1, layer.biases, 1);
-    for(i = 0; i < layer.outputs; ++i){
+    scal_cpu(layer.outputs, layer.momentum, layer.bias_updates, 1);
-        layer.bias_momentum[i] = layer.learning_rate*(layer.bias_updates[i]) + layer.momentum*layer.bias_momentum[i];
+
-        layer.biases[i] += layer.bias_momentum[i];
+    scal_cpu(layer.inputs*layer.outputs, 1.-layer.learning_rate*layer.decay, layer.weights, 1);
-    }
+    axpy_cpu(layer.inputs*layer.outputs, layer.learning_rate, layer.weight_updates, 1, layer.weights, 1);
-    for(i = 0; i < layer.outputs*layer.inputs; ++i){
+    scal_cpu(layer.inputs*layer.outputs, layer.momentum, layer.weight_updates, 1);
-        layer.weight_momentum[i] = layer.learning_rate*(layer.weight_updates[i] - layer.decay*layer.weights[i]) + layer.momentum*layer.weight_momentum[i];
-        layer.weights[i] += layer.weight_momentum[i];
-    }
-    memset(layer.bias_updates, 0, layer.outputs*sizeof(float));
-    memset(layer.weight_updates, 0, layer.outputs*layer.inputs*sizeof(float));
 }
 
 void forward_connected_layer(connected_layer layer, float *input)
 {
     int i;
     for(i = 0; i < layer.batch; ++i){
-        memcpy(layer.output+i*layer.outputs, layer.biases, layer.outputs*sizeof(float));
+        copy_cpu(layer.outputs, layer.biases, 1, layer.output + i*layer.outputs, 1);
     }
     int m = layer.batch;
     int k = layer.inputs;
@@ -82,8 +75,8 @@ void forward_connected_layer(connected_layer layer, float *input)
 void backward_connected_layer(connected_layer layer, float *input, float *delta)
 {
     int i;
+    gradient_array(layer.output, layer.outputs*layer.batch, layer.activation, layer.delta);
     for(i = 0; i < layer.outputs*layer.batch; ++i){
-        layer.delta[i] *= gradient(layer.output[i], layer.activation);
         layer.bias_updates[i%layer.outputs] += layer.delta[i];
     }
     int m = layer.inputs;

src/connected_layer.h

@@ -21,9 +21,6 @@ typedef struct{
     float *weight_adapt;
     float *bias_adapt;
 
-    float *weight_momentum;
-    float *bias_momentum;
-
     float *output;
     float *delta;
     

src/network.c

@@ -229,6 +229,8 @@ float *get_network_output_layer(network net, int i)
         return layer.output;
     } else if(net.types[i] == DROPOUT){
         return get_network_output_layer(net, i-1);
+    } else if(net.types[i] == FREEWEIGHT){
+        return get_network_output_layer(net, i-1);
     } else if(net.types[i] == CONNECTED){
         connected_layer layer = *(connected_layer *)net.layers[i];
         return layer.output;
@@ -258,6 +260,8 @@ float *get_network_delta_layer(network net, int i)
         return layer.delta;
     } else if(net.types[i] == DROPOUT){
         return get_network_delta_layer(net, i-1);
+    } else if(net.types[i] == FREEWEIGHT){
+        return get_network_delta_layer(net, i-1);
     } else if(net.types[i] == CONNECTED){
         connected_layer layer = *(connected_layer *)net.layers[i];
         return layer.delta;
@@ -424,6 +428,10 @@ int get_network_input_size_layer(network net, int i)
         dropout_layer layer = *(dropout_layer *) net.layers[i];
         return layer.inputs;
     }
+    else if(net.types[i] == FREEWEIGHT){
+        freeweight_layer layer = *(freeweight_layer *) net.layers[i];
+        return layer.inputs;
+    }
     else if(net.types[i] == SOFTMAX){
         softmax_layer layer = *(softmax_layer *)net.layers[i];
         return layer.inputs;
@@ -451,6 +459,10 @@ int get_network_output_size_layer(network net, int i)
         dropout_layer layer = *(dropout_layer *) net.layers[i];
         return layer.inputs;
     }
+    else if(net.types[i] == FREEWEIGHT){
+        freeweight_layer layer = *(freeweight_layer *) net.layers[i];
+        return layer.inputs;
+    }
     else if(net.types[i] == SOFTMAX){
         softmax_layer layer = *(softmax_layer *)net.layers[i];
         return layer.inputs;