From f047cfff99e00e28c02eb59b6d32386c122f9af6 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Sun, 08 Mar 2015 18:31:12 +0000
Subject: [PATCH] renamed sigmoid to logistic
---
src/connected_layer.c | 20 ++++++++++++--------
1 files changed, 12 insertions(+), 8 deletions(-)
diff --git a/src/connected_layer.c b/src/connected_layer.c
index 254d39e..642570c 100644
--- a/src/connected_layer.c
+++ b/src/connected_layer.c
@@ -36,7 +36,6 @@
float scale = 1./sqrt(inputs);
- //scale = .01;
for(i = 0; i < inputs*outputs; ++i){
layer->weights[i] = scale*rand_normal();
}
@@ -78,8 +77,6 @@
axpy_cpu(layer->outputs, 1, layer->bias_updates, 1, layer->bias_prev, 1);
scal_cpu(layer->outputs, 0, layer->bias_updates, 1);
- //printf("rate: %f\n", layer->learning_rate);
-
axpy_cpu(layer->outputs, layer->learning_rate, layer->bias_prev, 1, layer->biases, 1);
axpy_cpu(layer->inputs*layer->outputs, -layer->decay, layer->weights, 1, layer->weight_prev, 1);
@@ -115,9 +112,10 @@
void backward_connected_layer(connected_layer layer, float *input, float *delta)
{
int i;
+ float alpha = 1./layer.batch;
gradient_array(layer.output, layer.outputs*layer.batch, layer.activation, layer.delta);
for(i = 0; i < layer.batch; ++i){
- axpy_cpu(layer.outputs, 1, layer.delta + i*layer.outputs, 1, layer.bias_updates, 1);
+ axpy_cpu(layer.outputs, alpha, layer.delta + i*layer.outputs, 1, layer.bias_updates, 1);
}
int m = layer.inputs;
int k = layer.batch;
@@ -125,7 +123,7 @@
float *a = input;
float *b = layer.delta;
float *c = layer.weight_updates;
- gemm(1,0,m,n,k,1,a,m,b,n,1,c,n);
+ gemm(1,0,m,n,k,alpha,a,m,b,n,1,c,n);
m = layer.batch;
k = layer.outputs;
@@ -158,13 +156,18 @@
void update_connected_layer_gpu(connected_layer layer)
{
+/*
+ cuda_pull_array(layer.weights_gpu, layer.weights, layer.inputs*layer.outputs);
+ cuda_pull_array(layer.weight_updates_gpu, layer.weight_updates, layer.inputs*layer.outputs);
+ printf("Weights: %f updates: %f\n", mag_array(layer.weights, layer.inputs*layer.outputs), layer.learning_rate*mag_array(layer.weight_updates, layer.inputs*layer.outputs));
+*/
+
axpy_ongpu(layer.outputs, layer.learning_rate, layer.bias_updates_gpu, 1, layer.biases_gpu, 1);
scal_ongpu(layer.outputs, layer.momentum, layer.bias_updates_gpu, 1);
axpy_ongpu(layer.inputs*layer.outputs, -layer.decay, layer.weights_gpu, 1, layer.weight_updates_gpu, 1);
axpy_ongpu(layer.inputs*layer.outputs, layer.learning_rate, layer.weight_updates_gpu, 1, layer.weights_gpu, 1);
scal_ongpu(layer.inputs*layer.outputs, layer.momentum, layer.weight_updates_gpu, 1);
- //pull_connected_layer(layer);
}
void forward_connected_layer_gpu(connected_layer layer, float * input)
@@ -185,10 +188,11 @@
void backward_connected_layer_gpu(connected_layer layer, float * input, float * delta)
{
+ float alpha = 1./layer.batch;
int i;
gradient_array_ongpu(layer.output_gpu, layer.outputs*layer.batch, layer.activation, layer.delta_gpu);
for(i = 0; i < layer.batch; ++i){
- axpy_ongpu_offset(layer.outputs, 1, layer.delta_gpu, i*layer.outputs, 1, layer.bias_updates_gpu, 0, 1);
+ axpy_ongpu_offset(layer.outputs, alpha, layer.delta_gpu, i*layer.outputs, 1, layer.bias_updates_gpu, 0, 1);
}
int m = layer.inputs;
int k = layer.batch;
@@ -196,7 +200,7 @@
float * a = input;
float * b = layer.delta_gpu;
float * c = layer.weight_updates_gpu;
- gemm_ongpu(1,0,m,n,k,1,a,m,b,n,1,c,n);
+ gemm_ongpu(1,0,m,n,k,alpha,a,m,b,n,1,c,n);
m = layer.batch;
k = layer.outputs;
--
Gitblit v1.10.0