From f7a17f82eb43de864a4f980f235055da9685eef8 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Wed, 29 Jan 2014 00:28:42 +0000
Subject: [PATCH] Convolutional layers working w/ matrices
---
src/network.c | 172 +++++++++++++++++++++++++++++++++++++++++---------------
1 files changed, 125 insertions(+), 47 deletions(-)
diff --git a/src/network.c b/src/network.c
index cce673c..29e22e4 100644
--- a/src/network.c
+++ b/src/network.c
@@ -6,6 +6,7 @@
#include "connected_layer.h"
#include "convolutional_layer.h"
+//#include "old_conv.h"
#include "maxpool_layer.h"
#include "softmax_layer.h"
@@ -15,10 +16,12 @@
net.n = n;
net.layers = calloc(net.n, sizeof(void *));
net.types = calloc(net.n, sizeof(LAYER_TYPE));
+ net.outputs = 0;
+ net.output = 0;
return net;
}
-void forward_network(network net, double *input)
+void forward_network(network net, float *input)
{
int i;
for(i = 0; i < net.n; ++i){
@@ -45,13 +48,13 @@
}
}
-void update_network(network net, double step)
+void update_network(network net, float step, float momentum, float decay)
{
int i;
for(i = 0; i < net.n; ++i){
if(net.types[i] == CONVOLUTIONAL){
convolutional_layer layer = *(convolutional_layer *)net.layers[i];
- update_convolutional_layer(layer, step, 0.9, .01);
+ update_convolutional_layer(layer, step, momentum, decay);
}
else if(net.types[i] == MAXPOOL){
//maxpool_layer layer = *(maxpool_layer *)net.layers[i];
@@ -61,12 +64,12 @@
}
else if(net.types[i] == CONNECTED){
connected_layer layer = *(connected_layer *)net.layers[i];
- update_connected_layer(layer, step, .9, 0);
+ update_connected_layer(layer, step, momentum, 0);
}
}
}
-double *get_network_output_layer(network net, int i)
+float *get_network_output_layer(network net, int i)
{
if(net.types[i] == CONVOLUTIONAL){
convolutional_layer layer = *(convolutional_layer *)net.layers[i];
@@ -83,12 +86,12 @@
}
return 0;
}
-double *get_network_output(network net)
+float *get_network_output(network net)
{
return get_network_output_layer(net, net.n-1);
}
-double *get_network_delta_layer(network net, int i)
+float *get_network_delta_layer(network net, int i)
{
if(net.types[i] == CONVOLUTIONAL){
convolutional_layer layer = *(convolutional_layer *)net.layers[i];
@@ -106,16 +109,37 @@
return 0;
}
-double *get_network_delta(network net)
+float *get_network_delta(network net)
{
return get_network_delta_layer(net, net.n-1);
}
-void learn_network(network net, double *input)
+float calculate_error_network(network net, float *truth)
{
+ float sum = 0;
+ float *delta = get_network_delta(net);
+ float *out = get_network_output(net);
+ int i, k = get_network_output_size(net);
+ for(i = 0; i < k; ++i){
+ delta[i] = truth[i] - out[i];
+ sum += delta[i]*delta[i];
+ }
+ return sum;
+}
+
+int get_predicted_class_network(network net)
+{
+ float *out = get_network_output(net);
+ int k = get_network_output_size(net);
+ return max_index(out, k);
+}
+
+float backward_network(network net, float *input, float *truth)
+{
+ float error = calculate_error_network(net, truth);
int i;
- double *prev_input;
- double *prev_delta;
+ float *prev_input;
+ float *prev_delta;
for(i = net.n-1; i >= 0; --i){
if(i == 0){
prev_input = input;
@@ -126,8 +150,9 @@
}
if(net.types[i] == CONVOLUTIONAL){
convolutional_layer layer = *(convolutional_layer *)net.layers[i];
- learn_convolutional_layer(layer, prev_input);
- if(i != 0) backward_convolutional_layer(layer, prev_input, prev_delta);
+ learn_convolutional_layer(layer);
+ //learn_convolutional_layer(layer);
+ if(i != 0) backward_convolutional_layer(layer, prev_delta);
}
else if(net.types[i] == MAXPOOL){
maxpool_layer layer = *(maxpool_layer *)net.layers[i];
@@ -143,42 +168,65 @@
if(i != 0) backward_connected_layer(layer, prev_input, prev_delta);
}
}
+ return error;
}
-void train_network_batch(network net, batch b)
+float train_network_datum(network net, float *x, float *y, float step, float momentum, float decay)
{
- int i,j;
- int k = get_network_output_size(net);
+ forward_network(net, x);
+ int class = get_predicted_class_network(net);
+ float error = backward_network(net, x, y);
+ update_network(net, step, momentum, decay);
+ //return (y[class]?1:0);
+ return error;
+}
+
+float train_network_sgd(network net, data d, int n, float step, float momentum,float decay)
+{
+ int i;
+ float error = 0;
+ for(i = 0; i < n; ++i){
+ int index = rand()%d.X.rows;
+ error += train_network_datum(net, d.X.vals[index], d.y.vals[index], step, momentum, decay);
+ //if((i+1)%10 == 0){
+ // printf("%d: %f\n", (i+1), (float)correct/(i+1));
+ //}
+ }
+ return error/n;
+}
+float train_network_batch(network net, data d, int n, float step, float momentum,float decay)
+{
+ int i;
int correct = 0;
- for(i = 0; i < b.n; ++i){
- show_image(b.images[i], "Input");
- forward_network(net, b.images[i].data);
- image o = get_network_image(net);
- if(o.h) show_image_collapsed(o, "Output");
- double *output = get_network_output(net);
- double *delta = get_network_delta(net);
- int max_k = 0;
- double max = 0;
- for(j = 0; j < k; ++j){
- delta[j] = b.truth[i][j]-output[j];
- if(output[j] > max) {
- max = output[j];
- max_k = j;
- }
- }
- if(b.truth[i][max_k]) ++correct;
- printf("%f\n", (double)correct/(i+1));
- learn_network(net, b.images[i].data);
- update_network(net, .001);
+ for(i = 0; i < n; ++i){
+ int index = rand()%d.X.rows;
+ float *x = d.X.vals[index];
+ float *y = d.y.vals[index];
+ forward_network(net, x);
+ int class = get_predicted_class_network(net);
+ backward_network(net, x, y);
+ correct += (y[class]?1:0);
+ }
+ update_network(net, step, momentum, decay);
+ return (float)correct/n;
+
+}
+
+
+void train_network(network net, data d, float step, float momentum, float decay)
+{
+ int i;
+ int correct = 0;
+ for(i = 0; i < d.X.rows; ++i){
+ correct += train_network_datum(net, d.X.vals[i], d.y.vals[i], step, momentum, decay);
if(i%100 == 0){
visualize_network(net);
- cvWaitKey(100);
+ cvWaitKey(10);
}
}
visualize_network(net);
- print_network(net);
cvWaitKey(100);
- printf("Accuracy: %f\n", (double)correct/b.n);
+ printf("Accuracy: %f\n", (float)correct/d.X.rows);
}
int get_network_output_size_layer(network net, int i)
@@ -241,16 +289,37 @@
sprintf(buff, "Layer %d", i);
if(net.types[i] == CONVOLUTIONAL){
convolutional_layer layer = *(convolutional_layer *)net.layers[i];
- visualize_convolutional_filters(layer, buff);
+ visualize_convolutional_layer(layer, buff);
}
}
}
+float *network_predict(network net, float *input)
+{
+ forward_network(net, input);
+ float *out = get_network_output(net);
+ return out;
+}
+
+matrix network_predict_data(network net, data test)
+{
+ int i,j;
+ int k = get_network_output_size(net);
+ matrix pred = make_matrix(test.X.rows, k);
+ for(i = 0; i < test.X.rows; ++i){
+ float *out = network_predict(net, test.X.vals[i]);
+ for(j = 0; j < k; ++j){
+ pred.vals[i][j] = out[j];
+ }
+ }
+ return pred;
+}
+
void print_network(network net)
{
int i,j;
for(i = 0; i < net.n; ++i){
- double *output;
+ float *output = 0;
int n = 0;
if(net.types[i] == CONVOLUTIONAL){
convolutional_layer layer = *(convolutional_layer *)net.layers[i];
@@ -274,12 +343,21 @@
output = layer.output;
n = layer.inputs;
}
- double mean = mean_array(output, n);
- double vari = variance_array(output, n);
- printf("Layer %d - Mean: %f, Variance: %f\n",i,mean, vari);
+ float mean = mean_array(output, n);
+ float vari = variance_array(output, n);
+ fprintf(stderr, "Layer %d - Mean: %f, Variance: %f\n",i,mean, vari);
if(n > 100) n = 100;
- for(j = 0; j < n; ++j) printf("%f, ", output[j]);
- if(n == 100)printf(".....\n");
- printf("\n");
+ for(j = 0; j < n; ++j) fprintf(stderr, "%f, ", output[j]);
+ if(n == 100)fprintf(stderr,".....\n");
+ fprintf(stderr, "\n");
}
}
+
+float network_accuracy(network net, data d)
+{
+ matrix guess = network_predict_data(net, d);
+ float acc = matrix_accuracy(d.y, guess);
+ free_matrix(guess);
+ return acc;
+}
+
--
Gitblit v1.10.0