From a392bbd0c957a00e3782c96e7ced84a29ff9dd88 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Tue, 15 Mar 2016 05:33:02 +0000
Subject: [PATCH] Play along w/ alphago
---
src/softmax_layer.c | 44 +++++++++++++++++++++++---------------------
1 files changed, 23 insertions(+), 21 deletions(-)
diff --git a/src/softmax_layer.c b/src/softmax_layer.c
index e344d16..e189701 100644
--- a/src/softmax_layer.c
+++ b/src/softmax_layer.c
@@ -7,24 +7,26 @@
#include <stdio.h>
#include <assert.h>
-softmax_layer *make_softmax_layer(int batch, int inputs, int groups)
+softmax_layer make_softmax_layer(int batch, int inputs, int groups)
{
assert(inputs%groups == 0);
fprintf(stderr, "Softmax Layer: %d inputs\n", inputs);
- softmax_layer *layer = calloc(1, sizeof(softmax_layer));
- layer->batch = batch;
- layer->groups = groups;
- layer->inputs = inputs;
- layer->output = calloc(inputs*batch, sizeof(float));
- layer->delta = calloc(inputs*batch, sizeof(float));
+ softmax_layer l = {0};
+ l.type = SOFTMAX;
+ l.batch = batch;
+ l.groups = groups;
+ l.inputs = inputs;
+ l.outputs = inputs;
+ l.output = calloc(inputs*batch, sizeof(float));
+ l.delta = calloc(inputs*batch, sizeof(float));
#ifdef GPU
- layer->output_gpu = cuda_make_array(layer->output, inputs*batch);
- layer->delta_gpu = cuda_make_array(layer->delta, inputs*batch);
+ l.output_gpu = cuda_make_array(l.output, inputs*batch);
+ l.delta_gpu = cuda_make_array(l.delta, inputs*batch);
#endif
- return layer;
+ return l;
}
-void softmax_array(float *input, int n, float *output)
+void softmax_array(float *input, int n, float temp, float *output)
{
int i;
float sum = 0;
@@ -33,30 +35,30 @@
if(input[i] > largest) largest = input[i];
}
for(i = 0; i < n; ++i){
- sum += exp(input[i]-largest);
+ sum += exp(input[i]/temp-largest/temp);
}
- if(sum) sum = largest+log(sum);
+ if(sum) sum = largest/temp+log(sum);
else sum = largest-100;
for(i = 0; i < n; ++i){
- output[i] = exp(input[i]-sum);
+ output[i] = exp(input[i]/temp-sum);
}
}
-void forward_softmax_layer(const softmax_layer layer, network_state state)
+void forward_softmax_layer(const softmax_layer l, network_state state)
{
int b;
- int inputs = layer.inputs / layer.groups;
- int batch = layer.batch * layer.groups;
+ int inputs = l.inputs / l.groups;
+ int batch = l.batch * l.groups;
for(b = 0; b < batch; ++b){
- softmax_array(state.input+b*inputs, inputs, layer.output+b*inputs);
+ softmax_array(state.input+b*inputs, inputs, l.temperature, l.output+b*inputs);
}
}
-void backward_softmax_layer(const softmax_layer layer, network_state state)
+void backward_softmax_layer(const softmax_layer l, network_state state)
{
int i;
- for(i = 0; i < layer.inputs*layer.batch; ++i){
- state.delta[i] = layer.delta[i];
+ for(i = 0; i < l.inputs*l.batch; ++i){
+ state.delta[i] += l.delta[i];
}
}
--
Gitblit v1.10.0