FANN xor_train - eiichiromomma/CVMLAB GitHub Wiki
(FANN) xor_train
XORの学習のサンプル
#include <stdio.h>
#include "fann.h"
//コールバック関数のテスト
int FANN_API test_callback(struct fann *ann, struct fann_train_data *train,
unsigned int max_epochs, unsigned int epochs_between_reports,
float desired_error, unsigned int epochs)
{
printf("Epochs %8d. MSE: %.5f. Desired-MSE: %.5f\n", epochs, fann_get_MSE(ann), desired_error);
return 0;
}
int main()
{
//fann_typeは通常float。fanndouble.hを使うとdoubleになる。
fann_type *calc_out;
//ネットワークの構成パラメータ
const unsigned int num_input = 2;
const unsigned int num_output = 1;
const unsigned int num_layers = 3;
const unsigned int num_neurons_hidden = 3;
//学習パラメータ
const float desired_error = (const float) 0;
const unsigned int max_epochs = 1000;
const unsigned int epochs_between_reports = 10;
//ネットワーク構造体のポインタを作成
struct fann *ann;
//学習データのポインタを作成
struct fann_train_data *data;
unsigned int i = 0;
unsigned int decimal_point;
printf("Creating network.\n");
//ネットワークの作成
ann = fann_create_standard(num_layers, num_input, num_neurons_hidden, num_output);
//学習データの読み込み
data = fann_read_train_from_file("xor.data");
//傾きを1とするシグモイド関数(thresholdになる)
fann_set_activation_steepness_hidden(ann, 1);
fann_set_activation_steepness_output(ann, 1);
//Symmetricなシグモイド関数(データの範囲は(-1,1)
fann_set_activation_function_hidden(ann, FANN_SIGMOID_SYMMETRIC);
fann_set_activation_function_output(ann, FANN_SIGMOID_SYMMETRIC);
//終了判定を失敗ビット数にする
fann_set_train_stop_function(ann, FANN_STOPFUNC_BIT);
fann_set_bit_fail_limit(ann, 0.01f);
//重みの初期化
fann_init_weights(ann, data);
printf("Training network.\n");
//学習の開始
fann_train_on_data(ann, data, max_epochs, epochs_between_reports, desired_error);
//学習データを使ったテスト
printf("Testing network. %f\n", fann_test_data(ann, data));
//テスト結果の表示
for(i = 0; i < fann_length_train_data(data); i++)
{
calc_out = fann_run(ann, data->input[i]);
printf("XOR test (%f,%f) -> %f, should be %f, difference=%f\n",
data->input[i][0], data->input[i][1], calc_out[0], data->output[i][0],
fann_abs(calc_out[0] - data->output[i][0]));
}
printf("Saving network.\n");
//重みなどの保存
fann_save(ann, "xor_float.net");
//fixedの学習データを保存
decimal_point = fann_save_to_fixed(ann, "xor_fixed.net");
fann_save_train_to_fixed(data, "xor_fixed.data", decimal_point);
//後片付け
printf("Cleaning up.\n");
fann_destroy_train(data);
fann_destroy(ann);
return 0;
}
> ./xor_train
Creating network.
Training network.
Max epochs 1000. Desired error: 0.0000000000.
Epochs 1. Current error: 0.2710060775. Bit fail 4.
Epochs 10. Current error: 0.2684144974. Bit fail 4.
Epochs 20. Current error: 0.2478404641. Bit fail 4.
Epochs 30. Current error: 0.1754541397. Bit fail 4.
Epochs 40. Current error: 0.1287820041. Bit fail 4.
Epochs 50. Current error: 0.0897984356. Bit fail 4.
Epochs 60. Current error: 0.0565589033. Bit fail 4.
Epochs 70. Current error: 0.0363917761. Bit fail 4.
Epochs 80. Current error: 0.0257470459. Bit fail 4.
Epochs 90. Current error: 0.0164618175. Bit fail 4.
Epochs 100. Current error: 0.0116132954. Bit fail 4.
Epochs 110. Current error: 0.0067465315. Bit fail 4.
Epochs 120. Current error: 0.0047443812. Bit fail 4.
Epochs 130. Current error: 0.0029264868. Bit fail 4.
Epochs 140. Current error: 0.0018880134. Bit fail 4.
Epochs 150. Current error: 0.0012575694. Bit fail 4.
Epochs 160. Current error: 0.0007982652. Bit fail 4.
Epochs 170. Current error: 0.0005533139. Bit fail 4.
Epochs 180. Current error: 0.0003470994. Bit fail 4.
Epochs 190. Current error: 0.0002377222. Bit fail 4.
Epochs 200. Current error: 0.0001516260. Bit fail 2.
Epochs 210. Current error: 0.0000959772. Bit fail 1.
Epochs 220. Current error: 0.0000645029. Bit fail 1.
Epochs 227. Current error: 0.0000425604. Bit fail 0.
Testing network. 0.000038
XOR test (-1.000000,-1.000000) -> -0.982257, should be -1.000000, difference=0.017743
XOR test (-1.000000,1.000000) -> 0.990769, should be 1.000000, difference=0.009231
XOR test (1.000000,-1.000000) -> 0.991142, should be 1.000000, difference=0.008858
XOR test (1.000000,1.000000) -> -0.988301, should be -1.000000, difference=0.011699
Saving network.
Cleaning up.
たまたま上手く収束した。大抵1000回学習してギブアップする。