From ee25ad42c5e9ecdc5a3aa7125e657ce26cc9535c Mon Sep 17 00:00:00 2001
From: Edmond Yoo <hj3yoo@uwaterloo.ca>
Date: Sun, 16 Sep 2018 02:48:50 +0000
Subject: [PATCH] temp
---
src/connected_layer.c | 56 ++++++++++++++++++++++++++++++++++++++++++++++----------
1 files changed, 46 insertions(+), 10 deletions(-)
diff --git a/src/connected_layer.c b/src/connected_layer.c
index f20aa93..e6dc759 100644
--- a/src/connected_layer.c
+++ b/src/connected_layer.c
@@ -36,6 +36,10 @@
l.weights = calloc(outputs*inputs, sizeof(float));
l.biases = calloc(outputs, sizeof(float));
+ l.forward = forward_connected_layer;
+ l.backward = backward_connected_layer;
+ l.update = update_connected_layer;
+
//float scale = 1./sqrt(inputs);
float scale = sqrt(2./inputs);
for(i = 0; i < outputs*inputs; ++i){
@@ -66,6 +70,10 @@
}
#ifdef GPU
+ l.forward_gpu = forward_connected_layer_gpu;
+ l.backward_gpu = backward_connected_layer_gpu;
+ l.update_gpu = update_connected_layer_gpu;
+
l.weights_gpu = cuda_make_array(l.weights, outputs*inputs);
l.biases_gpu = cuda_make_array(l.biases, outputs);
@@ -89,10 +97,16 @@
l.x_gpu = cuda_make_array(l.output, l.batch*outputs);
l.x_norm_gpu = cuda_make_array(l.output, l.batch*outputs);
+#ifdef CUDNN
+ cudnnCreateTensorDescriptor(&l.normTensorDesc);
+ cudnnCreateTensorDescriptor(&l.dstTensorDesc);
+ cudnnSetTensor4dDescriptor(l.dstTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, l.batch, l.out_c, l.out_h, l.out_w);
+ cudnnSetTensor4dDescriptor(l.normTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, 1, l.out_c, 1, 1);
+#endif
}
#endif
l.activation = activation;
- fprintf(stderr, "Connected Layer: %d inputs, %d outputs\n", inputs, outputs);
+ fprintf(stderr, "connected %4d -> %4d\n", inputs, outputs);
return l;
}
@@ -187,14 +201,36 @@
{
int i, j;
for(i = 0; i < l.outputs; ++i){
- float scale = l.scales[i]/sqrt(l.rolling_variance[i] + .00001);
+ float scale = l.scales[i]/sqrt(l.rolling_variance[i] + .000001);
for(j = 0; j < l.inputs; ++j){
l.weights[i*l.inputs + j] *= scale;
}
l.biases[i] -= l.rolling_mean[i] * scale;
+ l.scales[i] = 1;
+ l.rolling_mean[i] = 0;
+ l.rolling_variance[i] = 1;
}
}
+
+void statistics_connected_layer(layer l)
+{
+ if(l.batch_normalize){
+ printf("Scales ");
+ print_statistics(l.scales, l.outputs);
+ /*
+ printf("Rolling Mean ");
+ print_statistics(l.rolling_mean, l.outputs);
+ printf("Rolling Variance ");
+ print_statistics(l.rolling_variance, l.outputs);
+ */
+ }
+ printf("Biases ");
+ print_statistics(l.biases, l.outputs);
+ printf("Weights ");
+ print_statistics(l.weights, l.outputs);
+}
+
#ifdef GPU
void pull_connected_layer(connected_layer l)
@@ -250,20 +286,20 @@
float * b = l.weights_gpu;
float * c = l.output_gpu;
gemm_ongpu(0,1,m,n,k,1,a,k,b,k,1,c,n);
- if(l.batch_normalize){
- forward_batchnorm_layer_gpu(l, state);
- }
- for(i = 0; i < l.batch; ++i){
- axpy_ongpu(l.outputs, 1, l.biases_gpu, 1, l.output_gpu + i*l.outputs, 1);
- }
+ if (l.batch_normalize) {
+ forward_batchnorm_layer_gpu(l, state);
+ }
+ else {
+ add_bias_gpu(l.output_gpu, l.biases_gpu, l.batch, l.outputs, 1);
+ }
+ //for(i = 0; i < l.batch; ++i) axpy_ongpu(l.outputs, 1, l.biases_gpu, 1, l.output_gpu + i*l.outputs, 1);
activate_array_ongpu(l.output_gpu, l.outputs*l.batch, l.activation);
-
}
void backward_connected_layer_gpu(connected_layer l, network_state state)
{
int i;
- constrain_ongpu(l.outputs*l.batch, 5, l.delta_gpu, 1);
+ constrain_ongpu(l.outputs*l.batch, 1, l.delta_gpu, 1);
gradient_array_ongpu(l.output_gpu, l.outputs*l.batch, l.activation, l.delta_gpu);
for(i = 0; i < l.batch; ++i){
axpy_ongpu(l.outputs, 1, l.delta_gpu + i*l.outputs, 1, l.bias_updates_gpu, 1);
--
Gitblit v1.10.0