Working?

Joseph Redmon

2013-12-03 0d6bb5d44d8e815ebf6ccce1dae2f83178780e7b

 src/connected_layer.c

@@ -1,43 +1,41 @@
#include "connected_layer.h"
#include "utils.h"

#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

connected_layer make_connected_layer(int inputs, int outputs, ACTIVATOR_TYPE activator)
connected_layer *make_connected_layer(int inputs, int outputs, ACTIVATION activation)
{
    printf("Connected Layer: %d inputs, %d outputs\n", inputs, outputs);
    int i;
    connected_layer layer;
    layer.inputs = inputs;
    layer.outputs = outputs;
    connected_layer *layer = calloc(1, sizeof(connected_layer));
    layer->inputs = inputs;
    layer->outputs = outputs;

    layer.output = calloc(outputs, sizeof(double*));
    layer->output = calloc(outputs, sizeof(double*));
    layer->delta = calloc(outputs, sizeof(double*));

    layer.weight_updates = calloc(inputs*outputs, sizeof(double));
    layer.weights = calloc(inputs*outputs, sizeof(double));
    layer->weight_updates = calloc(inputs*outputs, sizeof(double));
    layer->weight_momentum = calloc(inputs*outputs, sizeof(double));
    layer->weights = calloc(inputs*outputs, sizeof(double));
    double scale = 2./inputs;
    for(i = 0; i < inputs*outputs; ++i)
        layer.weights[i] = .5 - (double)rand()/RAND_MAX;
        layer->weights[i] = rand_normal()*scale;

    layer.bias_updates = calloc(outputs, sizeof(double));
    layer.biases = calloc(outputs, sizeof(double));
    layer->bias_updates = calloc(outputs, sizeof(double));
    layer->bias_momentum = calloc(outputs, sizeof(double));
    layer->biases = calloc(outputs, sizeof(double));
    for(i = 0; i < outputs; ++i)
        layer.biases[i] = (double)rand()/RAND_MAX;
        //layer->biases[i] = rand_normal()*scale + scale;
        layer->biases[i] = 1;

    if(activator == SIGMOID){
        layer.activation = sigmoid_activation;
        layer.gradient = sigmoid_gradient;
    }else if(activator == RELU){
        layer.activation = relu_activation;
        layer.gradient = relu_gradient;
    }else if(activator == IDENTITY){
        layer.activation = identity_activation;
        layer.gradient = identity_gradient;
    }

    layer->activation = activation;
    return layer;
}

void run_connected_layer(double *input, connected_layer layer)
void forward_connected_layer(connected_layer layer, double *input)
{
    int i, j;
    for(i = 0; i < layer.outputs; ++i){
@@ -45,52 +43,48 @@
        for(j = 0; j < layer.inputs; ++j){
            layer.output[i] += input[j]*layer.weights[i*layer.inputs + j];
        }
        layer.output[i] = layer.activation(layer.output[i]);
        layer.output[i] = activate(layer.output[i], layer.activation);
    }
}

void learn_connected_layer(double *input, connected_layer layer)
void learn_connected_layer(connected_layer layer, double *input)
{
    calculate_update_connected_layer(input, layer);
    backpropagate_connected_layer(input, layer);
    int i, j;
    for(i = 0; i < layer.outputs; ++i){
        layer.delta[i] *= gradient(layer.output[i], layer.activation);
        layer.bias_updates[i] += layer.delta[i];
        for(j = 0; j < layer.inputs; ++j){
            layer.weight_updates[i*layer.inputs + j] += layer.delta[i]*input[j];
        }
    }
}

void update_connected_layer(connected_layer layer, double step)
void update_connected_layer(connected_layer layer, double step, double momentum, double decay)
{
    int i,j;
    for(i = 0; i < layer.outputs; ++i){
        layer.biases[i] += step*layer.bias_updates[i];
        layer.bias_momentum[i] = step*(layer.bias_updates[i]) + momentum*layer.bias_momentum[i];
        layer.biases[i] += layer.bias_momentum[i];
        for(j = 0; j < layer.inputs; ++j){
            int index = i*layer.inputs+j;
            layer.weights[index] += step*layer.weight_updates[index];
            layer.weight_momentum[index] = step*(layer.weight_updates[index] - decay*layer.weights[index]) + momentum*layer.weight_momentum[index];
            layer.weights[index] += layer.weight_momentum[index];
            //layer.weights[index] = constrain(layer.weights[index], 100.);
        }
    }
    memset(layer.bias_updates, 0, layer.outputs*sizeof(double));
    memset(layer.weight_updates, 0, layer.outputs*layer.inputs*sizeof(double));
}

void calculate_update_connected_layer(double *input, connected_layer layer)
{
    int i, j;
    for(i = 0; i < layer.outputs; ++i){
        layer.bias_updates[i] += layer.output[i];
        for(j = 0; j < layer.inputs; ++j){
            layer.weight_updates[i*layer.inputs + j] += layer.output[i]*input[j];
        }
    }
}

void backpropagate_connected_layer(double *input, connected_layer layer)
void backward_connected_layer(connected_layer layer, double *input, double *delta)
{
    int i, j;

    for(j = 0; j < layer.inputs; ++j){
        double grad = layer.gradient(input[j]);
        input[j] = 0;
        delta[j] = 0;
        for(i = 0; i < layer.outputs; ++i){
            input[j] += layer.output[i]*layer.weights[i*layer.inputs + j];
            delta[j] += layer.delta[i]*layer.weights[i*layer.inputs + j];
        }
        input[j] *= grad;
    }
}