From cd8a3dcb4ca42f22ad8f46a95e00977c92be6bbd Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Thu, 08 Feb 2018 23:22:42 +0000
Subject: [PATCH] Compile fixes
---
src/network.c | 305 +++++++++++++++++++++++++++++++++++---------------
1 files changed, 213 insertions(+), 92 deletions(-)
diff --git a/src/network.c b/src/network.c
index 68790e5..0c1b9af 100644
--- a/src/network.c
+++ b/src/network.c
@@ -1,36 +1,116 @@
#include <stdio.h>
#include <time.h>
+#include <assert.h>
#include "network.h"
#include "image.h"
#include "data.h"
#include "utils.h"
+#include "blas.h"
#include "crop_layer.h"
#include "connected_layer.h"
+#include "gru_layer.h"
+#include "rnn_layer.h"
+#include "crnn_layer.h"
+#include "local_layer.h"
#include "convolutional_layer.h"
-#include "deconvolutional_layer.h"
+#include "activation_layer.h"
#include "detection_layer.h"
+#include "region_layer.h"
+#include "normalization_layer.h"
+#include "batchnorm_layer.h"
#include "maxpool_layer.h"
+#include "reorg_layer.h"
+#include "avgpool_layer.h"
#include "cost_layer.h"
#include "softmax_layer.h"
#include "dropout_layer.h"
#include "route_layer.h"
+#include "shortcut_layer.h"
+
+int get_current_batch(network net)
+{
+ int batch_num = (*net.seen)/(net.batch*net.subdivisions);
+ return batch_num;
+}
+
+void reset_momentum(network net)
+{
+ if (net.momentum == 0) return;
+ net.learning_rate = 0;
+ net.momentum = 0;
+ net.decay = 0;
+ #ifdef GPU
+ //if(net.gpu_index >= 0) update_network_gpu(net);
+ #endif
+}
+
+float get_current_rate(network net)
+{
+ int batch_num = get_current_batch(net);
+ int i;
+ float rate;
+ if (batch_num < net.burn_in) return net.learning_rate * pow((float)batch_num / net.burn_in, net.power);
+ switch (net.policy) {
+ case CONSTANT:
+ return net.learning_rate;
+ case STEP:
+ return net.learning_rate * pow(net.scale, batch_num/net.step);
+ case STEPS:
+ rate = net.learning_rate;
+ for(i = 0; i < net.num_steps; ++i){
+ if(net.steps[i] > batch_num) return rate;
+ rate *= net.scales[i];
+ //if(net.steps[i] > batch_num - 1 && net.scales[i] > 1) reset_momentum(net);
+ }
+ return rate;
+ case EXP:
+ return net.learning_rate * pow(net.gamma, batch_num);
+ case POLY:
+ return net.learning_rate * pow(1 - (float)batch_num / net.max_batches, net.power);
+ //if (batch_num < net.burn_in) return net.learning_rate * pow((float)batch_num / net.burn_in, net.power);
+ //return net.learning_rate * pow(1 - (float)batch_num / net.max_batches, net.power);
+ case RANDOM:
+ return net.learning_rate * pow(rand_uniform(0,1), net.power);
+ case SIG:
+ return net.learning_rate * (1./(1.+exp(net.gamma*(batch_num - net.step))));
+ default:
+ fprintf(stderr, "Policy is weird!\n");
+ return net.learning_rate;
+ }
+}
char *get_layer_string(LAYER_TYPE a)
{
switch(a){
case CONVOLUTIONAL:
return "convolutional";
+ case ACTIVE:
+ return "activation";
+ case LOCAL:
+ return "local";
case DECONVOLUTIONAL:
return "deconvolutional";
case CONNECTED:
return "connected";
+ case RNN:
+ return "rnn";
+ case GRU:
+ return "gru";
+ case CRNN:
+ return "crnn";
case MAXPOOL:
return "maxpool";
+ case REORG:
+ return "reorg";
+ case AVGPOOL:
+ return "avgpool";
case SOFTMAX:
return "softmax";
case DETECTION:
return "detection";
+ case REGION:
+ return "region";
case DROPOUT:
return "dropout";
case CROP:
@@ -39,6 +119,12 @@
return "cost";
case ROUTE:
return "route";
+ case SHORTCUT:
+ return "shortcut";
+ case NORMALIZATION:
+ return "normalization";
+ case BATCHNORM:
+ return "batchnorm";
default:
break;
}
@@ -50,6 +136,7 @@
network net = {0};
net.n = n;
net.layers = calloc(net.n, sizeof(layer));
+ net.seen = calloc(1, sizeof(int));
#ifdef GPU
net.input_gpu = calloc(1, sizeof(float *));
net.truth_gpu = calloc(1, sizeof(float *));
@@ -59,30 +146,15 @@
void forward_network(network net, network_state state)
{
+ state.workspace = net.workspace;
int i;
for(i = 0; i < net.n; ++i){
+ state.index = i;
layer l = net.layers[i];
- if(l.type == CONVOLUTIONAL){
- forward_convolutional_layer(l, state);
- } else if(l.type == DECONVOLUTIONAL){
- forward_deconvolutional_layer(l, state);
- } else if(l.type == DETECTION){
- forward_detection_layer(l, state);
- } else if(l.type == CONNECTED){
- forward_connected_layer(l, state);
- } else if(l.type == CROP){
- forward_crop_layer(l, state);
- } else if(l.type == COST){
- forward_cost_layer(l, state);
- } else if(l.type == SOFTMAX){
- forward_softmax_layer(l, state);
- } else if(l.type == MAXPOOL){
- forward_maxpool_layer(l, state);
- } else if(l.type == DROPOUT){
- forward_dropout_layer(l, state);
- } else if(l.type == ROUTE){
- forward_route_layer(l, net);
+ if(l.delta){
+ scal_cpu(l.outputs * l.batch, 0, l.delta, 1);
}
+ l.forward(l, state);
state.input = l.output;
}
}
@@ -91,20 +163,20 @@
{
int i;
int update_batch = net.batch*net.subdivisions;
+ float rate = get_current_rate(net);
for(i = 0; i < net.n; ++i){
layer l = net.layers[i];
- if(l.type == CONVOLUTIONAL){
- update_convolutional_layer(l, update_batch, net.learning_rate, net.momentum, net.decay);
- } else if(l.type == DECONVOLUTIONAL){
- update_deconvolutional_layer(l, net.learning_rate, net.momentum, net.decay);
- } else if(l.type == CONNECTED){
- update_connected_layer(l, update_batch, net.learning_rate, net.momentum, net.decay);
+ if(l.update){
+ l.update(l, update_batch, rate, net.momentum, net.decay);
}
}
}
float *get_network_output(network net)
{
+#ifdef GPU
+ if (gpu_index >= 0) return get_network_output_gpu(net);
+#endif
int i;
for(i = net.n-1; i > 0; --i) if(net.layers[i].type != COST) break;
return net.layers[i].output;
@@ -112,13 +184,16 @@
float get_network_cost(network net)
{
- if(net.layers[net.n-1].type == COST){
- return net.layers[net.n-1].output[0];
+ int i;
+ float sum = 0;
+ int count = 0;
+ for(i = 0; i < net.n; ++i){
+ if(net.layers[i].cost){
+ sum += net.layers[i].cost[0];
+ ++count;
+ }
}
- if(net.layers[net.n-1].type == DETECTION){
- return net.layers[net.n-1].cost[0];
- }
- return 0;
+ return sum/count;
}
int get_predicted_class_network(network net)
@@ -132,51 +207,41 @@
{
int i;
float *original_input = state.input;
+ float *original_delta = state.delta;
+ state.workspace = net.workspace;
for(i = net.n-1; i >= 0; --i){
+ state.index = i;
if(i == 0){
state.input = original_input;
- state.delta = 0;
+ state.delta = original_delta;
}else{
layer prev = net.layers[i-1];
state.input = prev.output;
state.delta = prev.delta;
}
layer l = net.layers[i];
- if(l.type == CONVOLUTIONAL){
- backward_convolutional_layer(l, state);
- } else if(l.type == DECONVOLUTIONAL){
- backward_deconvolutional_layer(l, state);
- } else if(l.type == MAXPOOL){
- if(i != 0) backward_maxpool_layer(l, state);
- } else if(l.type == DROPOUT){
- backward_dropout_layer(l, state);
- } else if(l.type == DETECTION){
- backward_detection_layer(l, state);
- } else if(l.type == SOFTMAX){
- if(i != 0) backward_softmax_layer(l, state);
- } else if(l.type == CONNECTED){
- backward_connected_layer(l, state);
- } else if(l.type == COST){
- backward_cost_layer(l, state);
- } else if(l.type == ROUTE){
- backward_route_layer(l, net);
- }
+ if (l.stopbackward) break;
+ l.backward(l, state);
}
}
float train_network_datum(network net, float *x, float *y)
{
- #ifdef GPU
+#ifdef GPU
if(gpu_index >= 0) return train_network_datum_gpu(net, x, y);
- #endif
+#endif
network_state state;
+ *net.seen += net.batch;
+ state.index = 0;
+ state.net = net;
state.input = x;
+ state.delta = 0;
state.truth = y;
state.train = 1;
forward_network(net, state);
backward_network(net, state);
float error = get_network_cost(net);
- if((net.seen/net.batch)%net.subdivisions == 0) update_network(net);
+ if(((*net.seen)/net.batch)%net.subdivisions == 0) update_network(net);
return error;
}
@@ -189,7 +254,6 @@
int i;
float sum = 0;
for(i = 0; i < n; ++i){
- net.seen += batch;
get_random_batch(d, batch, X, y);
float err = train_network_datum(net, X, y);
sum += err;
@@ -201,6 +265,7 @@
float train_network(network net, data d)
{
+ assert(d.X.rows % net.batch == 0);
int batch = net.batch;
int n = d.X.rows / batch;
float *X = calloc(batch*d.X.cols, sizeof(float));
@@ -210,7 +275,6 @@
float sum = 0;
for(i = 0; i < n; ++i){
get_next_batch(d, batch, i*batch, X, y);
- net.seen += batch;
float err = train_network_datum(net, X, y);
sum += err;
}
@@ -219,11 +283,15 @@
return (float)sum/(n*batch);
}
+
float train_network_batch(network net, data d, int n)
{
int i,j;
network_state state;
+ state.index = 0;
+ state.net = net;
state.train = 1;
+ state.delta = 0;
float sum = 0;
int batch = 2;
for(i = 0; i < n; ++i){
@@ -246,46 +314,81 @@
int i;
for(i = 0; i < net->n; ++i){
net->layers[i].batch = b;
+#ifdef CUDNN
+ if(net->layers[i].type == CONVOLUTIONAL){
+ cudnn_convolutional_setup(net->layers + i);
+ }
+#endif
}
}
-/*
-int resize_network(network net, int h, int w, int c)
+int resize_network(network *net, int w, int h)
{
- fprintf(stderr, "Might be broken, careful!!");
+#ifdef GPU
+ cuda_set_device(net->gpu_index);
+ if(gpu_index >= 0){
+ cuda_free(net->workspace);
+ }
+#endif
int i;
- for (i = 0; i < net.n; ++i){
- if(net.types[i] == CONVOLUTIONAL){
- convolutional_layer *layer = (convolutional_layer *)net.layers[i];
- resize_convolutional_layer(layer, h, w);
- image output = get_convolutional_image(*layer);
- h = output.h;
- w = output.w;
- c = output.c;
- } else if(net.types[i] == DECONVOLUTIONAL){
- deconvolutional_layer *layer = (deconvolutional_layer *)net.layers[i];
- resize_deconvolutional_layer(layer, h, w);
- image output = get_deconvolutional_image(*layer);
- h = output.h;
- w = output.w;
- c = output.c;
- }else if(net.types[i] == MAXPOOL){
- maxpool_layer *layer = (maxpool_layer *)net.layers[i];
- resize_maxpool_layer(layer, h, w);
- image output = get_maxpool_image(*layer);
- h = output.h;
- w = output.w;
- c = output.c;
- }else if(net.types[i] == DROPOUT){
- dropout_layer *layer = (dropout_layer *)net.layers[i];
- resize_dropout_layer(layer, h*w*c);
+ //if(w == net->w && h == net->h) return 0;
+ net->w = w;
+ net->h = h;
+ int inputs = 0;
+ size_t workspace_size = 0;
+ //fprintf(stderr, "Resizing to %d x %d...\n", w, h);
+ //fflush(stderr);
+ for (i = 0; i < net->n; ++i){
+ layer l = net->layers[i];
+ if(l.type == CONVOLUTIONAL){
+ resize_convolutional_layer(&l, w, h);
+ }else if(l.type == CROP){
+ resize_crop_layer(&l, w, h);
+ }else if(l.type == MAXPOOL){
+ resize_maxpool_layer(&l, w, h);
+ }else if(l.type == REGION){
+ resize_region_layer(&l, w, h);
+ }else if(l.type == ROUTE){
+ resize_route_layer(&l, net);
+ }else if(l.type == REORG){
+ resize_reorg_layer(&l, w, h);
+ }else if(l.type == AVGPOOL){
+ resize_avgpool_layer(&l, w, h);
+ }else if(l.type == NORMALIZATION){
+ resize_normalization_layer(&l, w, h);
+ }else if(l.type == COST){
+ resize_cost_layer(&l, inputs);
}else{
+ fprintf(stderr, "Resizing type %d \n", (int)l.type);
error("Cannot resize this type of layer");
}
+ if(l.workspace_size > workspace_size) workspace_size = l.workspace_size;
+ inputs = l.outputs;
+ net->layers[i] = l;
+ w = l.out_w;
+ h = l.out_h;
+ if(l.type == AVGPOOL) break;
}
+#ifdef GPU
+ if(gpu_index >= 0){
+ if(net->input_gpu) {
+ cuda_free(*net->input_gpu);
+ *net->input_gpu = 0;
+ cuda_free(*net->truth_gpu);
+ *net->truth_gpu = 0;
+ }
+ net->workspace = cuda_make_array(0, (workspace_size-1)/sizeof(float)+1);
+ }else {
+ free(net->workspace);
+ net->workspace = calloc(1, workspace_size);
+ }
+#else
+ free(net->workspace);
+ net->workspace = calloc(1, workspace_size);
+#endif
+ //fprintf(stderr, " Done!\n");
return 0;
}
-*/
int get_network_output_size(network net)
{
@@ -362,6 +465,8 @@
#endif
network_state state;
+ state.net = net;
+ state.index = 0;
state.input = input;
state.truth = 0;
state.train = 0;
@@ -469,17 +574,16 @@
return acc;
}
-float *network_accuracies(network net, data d)
+float *network_accuracies(network net, data d, int n)
{
static float acc[2];
matrix guess = network_predict_data(net, d);
- acc[0] = matrix_topk_accuracy(d.y, guess,1);
- acc[1] = matrix_topk_accuracy(d.y, guess,5);
+ acc[0] = matrix_topk_accuracy(d.y, guess, 1);
+ acc[1] = matrix_topk_accuracy(d.y, guess, n);
free_matrix(guess);
return acc;
}
-
float network_accuracy_multi(network net, data d, int n)
{
matrix guess = network_predict_data_multi(net, d, n);
@@ -488,4 +592,21 @@
return acc;
}
-
+void free_network(network net)
+{
+ int i;
+ for (i = 0; i < net.n; ++i) {
+ free_layer(net.layers[i]);
+ }
+ free(net.layers);
+#ifdef GPU
+ if (gpu_index >= 0) cuda_free(net.workspace);
+ else free(net.workspace);
+ if (*net.input_gpu) cuda_free(*net.input_gpu);
+ if (*net.truth_gpu) cuda_free(*net.truth_gpu);
+ if (net.input_gpu) free(net.input_gpu);
+ if (net.truth_gpu) free(net.truth_gpu);
+#else
+ free(net.workspace);
+#endif
+}
--
Gitblit v1.10.0