FANN cascade_train - eiichiromomma/CVMLAB GitHub Wiki
(FANN) cascade_train
カスケード学習の例
//カスケード学習の例
#include <stdio.h>
#include "fann.h"
int main()
{
//ネットワークのポインタ
struct fann *ann;
//学習とテストデータのポインタ
struct fann_train_data *train_data, *test_data;
//終了条件
const float desired_error = (const float) 0.001;
//Cascade2で増加させるニューロンの最大数
unsigned int max_neurons = 40;
//1ニューロン増加させるごとにレポート
unsigned int neurons_between_reports = 1;
printf("Reading data.\n");
//学習とテストデータの読み込み
//(fann-2.0のソースファイルに含まれるので必要に応じてパスを書き換える)
train_data = fann_read_train_from_file("../benchmarks/datasets/two-spiral.train");
test_data = fann_read_train_from_file("../benchmarks/datasets/two-spiral.test");
//学習とテストデータの正規化
fann_scale_train_data(train_data, 0, 1);
fann_scale_train_data(test_data, 0, 1);
printf("Creating network.\n");
//ショートカット接続のネットワークを作成
ann = fann_create_shortcut(2, fann_num_input_train_data(train_data), fann_num_output_train_data(train_data));
//学習アルゴリズムの設定
fann_set_training_algorithm(ann, FANN_TRAIN_RPROP);
//関数の設定
fann_set_activation_function_hidden(ann, FANN_SIGMOID_SYMMETRIC);
fann_set_activation_function_output(ann, FANN_LINEAR_PIECE);
fann_set_train_error_function(ann, FANN_ERRORFUNC_LINEAR);
//ネットワークパラメータの表示
fann_print_parameters(ann);
printf("Training network.\n");
//カスケード学習の開始
fann_cascadetrain_on_data(ann, train_data, max_neurons, neurons_between_reports, desired_error);
//ネットワーク接続状態の表示
fann_print_connections(ann);
//テスト結果の表示
printf("\nTrain error: %f, Test error: %f\n\n", fann_test_data(ann, train_data),
fann_test_data(ann, test_data));
printf("Saving network.\n");
//ネットワークの保存
fann_save(ann, "two_spirali.net");
//後片付け
printf("Cleaning up.\n");
fann_destroy_train(train_data);
fann_destroy_train(test_data);
fann_destroy(ann);
return 0;
}192 2 1
0.296027 0.703973
0
0.385146 0.671890
1
0.401172 0.352095
0
みたいな192個の入力2出力1のデータ
Input layer : 2 neurons, 1 bias
Output layer : 1 neurons
Total neurons and biases : 4
Total connections : 3
Connection rate : 1.000
Shortcut connections : 1
Training algorithm : FANN_TRAIN_RPROP
Training error function : FANN_ERRORFUNC_LINEAR
Training stop function : FANN_STOPFUNC_MSE
Learning rate : 0.700
Learning momentum : 0.000
Quickprop decay : -0.000100
Quickprop mu : 1.750
RPROP increase factor : 1.200
RPROP decrease factor : 0.500
RPROP delta min : 0.000
RPROP delta max : 50.000
Cascade output change fraction : 0.010000
Cascade candidate change fraction : 0.010000
Cascade output stagnation epochs : 12
Cascade candidate stagnation epochs : 12
Cascade max output epochs : 150
Cascade max candidate epochs : 150
Cascade weight multiplier : 0.400
Cascade candidate limit :1000.000
Cascade activation functions[0] : FANN_SIGMOID
Cascade activation functions[1] : FANN_SIGMOID_SYMMETRIC
Cascade activation functions[2] : FANN_GAUSSIAN
Cascade activation functions[3] : FANN_GAUSSIAN_SYMMETRIC
Cascade activation functions[4] : FANN_ELLIOT
Cascade activation functions[5] : FANN_ELLIOT_SYMMETRIC
Cascade activation steepnesses[0] : 0.250
Cascade activation steepnesses[1] : 0.500
Cascade activation steepnesses[2] : 0.750
Cascade activation steepnesses[3] : 1.000
Cascade candidate groups : 2
Cascade no. of candidates : 48
Cascade2をやる前に表示したので中間層が無いのに注目
Layer / Neuron 012345678901234567890123456789012
L 1 / N 3 ZeD..............................
L 2 / N 4 jBAi.............................
L 3 / N 5 ezJZZ............................
L 4 / N 6 ddbeAZ...........................
L 5 / N 7 ddAzZhh..........................
L 6 / N 8 fdAzKysd.........................
L 7 / N 9 AAaABAAAB........................
L 8 / N 10 bccFMfIZeZ.......................
L 9 / N 11 caAaEBaCCbC......................
L 10 / N 12 CcaZzBlbzzLo.....................
L 11 / N 13 aABcCCbbglbBA....................
L 12 / N 14 albcTjcazBleBB...................
L 13 / N 15 BBBBZZRKbcACaEg..................
L 14 / N 16 BbACEAbMqKaAabFC.................
L 15 / N 17 BbaBcABdSfbbbAnCc................
L 16 / N 18 afbzzczQzkAbCBzCBB...............
L 17 / N 19 aAaacbABBfbAAAgbbbC..............
L 18 / N 20 AaabHbaEmaaaBbedCaHc.............
L 19 / N 21 abARZbEOGBBDbBzBBCNaC............
L 20 / N 22 afaCzczCUfjAaAzzBbaaac...........
L 21 / N 23 AABJECbAGZMaBJZGAACbANJ..........
L 22 / N 24 BdazzezzzMbaACpzaafbazCE.........
L 23 / N 25 baazbzzzbbzaAbzDAczEazzdn........
L 24 / N 26 cbazhdzcdlbAaazCACdBbzbcqD.......
L 25 / N 27 cbazhAzzaBkbBCzzaBzaBzzBazL......
L 26 / N 28 BBBCdCzLFZCbCDWcaBIbcBzBSbZA.....
L 27 / N 29 acabzzzZOdzCabzzCBGCabzzzeDDz....
L 28 / N 30 cBAqgzamazzABazEbAzbamBafuzzBz...
L 29 / N 31 AaAZZOECdAcAaAxCbCEfBhDbCbcgCCZ..
L 30 / N 32 AfaIzbzfAmyaABdFaAybCbzCwMFckzza.
L 31 / N 33 BcbzFmZzDZeceBZHcBOdbMuDPPOkGeZcI
Train error: 0.000021, Test error: 0.113035
イメージが湧かないが要するに全部が並列に繋がっているということだろう。
Saving network.
Input layer : 2 neurons, 1 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Hidden layer : 1 neurons, 0 bias
Output layer : 1 neurons
Total neurons and biases : 34
Total connections : 558
Connection rate : 1.000
Shortcut connections : 1
Training algorithm : FANN_TRAIN_RPROP
Training error function : FANN_ERRORFUNC_LINEAR
Training stop function : FANN_STOPFUNC_MSE
Learning rate : 0.700
Learning momentum : 0.000
Quickprop decay : -0.000100
Quickprop mu : 1.750
RPROP increase factor : 1.200
RPROP decrease factor : 0.500
RPROP delta min : 0.000
RPROP delta max : 50.000
Cascade output change fraction : 0.010000
Cascade candidate change fraction : 0.010000
Cascade output stagnation epochs : 12
Cascade candidate stagnation epochs : 12
Cascade max output epochs : 150
Cascade max candidate epochs : 150
Cascade weight multiplier : 0.400
Cascade candidate limit :1000.000
Cascade activation functions[0] : FANN_SIGMOID
Cascade activation functions[1] : FANN_SIGMOID_SYMMETRIC
Cascade activation functions[2] : FANN_GAUSSIAN
Cascade activation functions[3] : FANN_GAUSSIAN_SYMMETRIC
Cascade activation functions[4] : FANN_ELLIOT
Cascade activation functions[5] : FANN_ELLIOT_SYMMETRIC
Cascade activation steepnesses[0] : 0.250
Cascade activation steepnesses[1] : 0.500
Cascade activation steepnesses[2] : 0.750
Cascade activation steepnesses[3] : 1.000
Cascade candidate groups : 2
Cascade no. of candidates : 48
Cascadeなので1のバイアス無しが続く。
Training network.
Max neurons 40. Desired error: 0.001000
Neurons 1. Current error: 0.247635. Total error: 47.5459. Epochs 56. Bit fail 192
Neurons 2. Current error: 0.242957. Total error: 46.6477. Epochs 320. Bit fail 186. candidate steepness 0.50. function FANN_GAUSSIAN
Neurons 3. Current error: 0.240952. Total error: 46.2627. Epochs 574. Bit fail 186. candidate steepness 1.00. function FANN_SIGMOID
Neurons 4. Current error: 0.232480. Total error: 44.6362. Epochs 775. Bit fail 176. candidate steepness 1.00. function FANN_ELLIOT
Neurons 5. Current error: 0.221806. Total error: 42.5867. Epochs 961. Bit fail 171. candidate steepness 1.00. function FANN_GAUSSIAN
Neurons 6. Current error: 0.213415. Total error: 40.9758. Epochs 1204. Bit fail 161. candidate steepness 1.00. function FANN_SIGMOID
Neurons 7. Current error: 0.205998. Total error: 39.5516. Epochs 1504. Bit fail 153. candidate steepness 0.50. function FANN_GAUSSIAN
Neurons 8. Current error: 0.204457. Total error: 39.2557. Epochs 1713. Bit fail 143. candidate steepness 1.00. function FANN_SIGMOID_SYMMETRIC
Neurons 9. Current error: 0.192103. Total error: 36.8839. Epochs 1904. Bit fail 126. candidate steepness 0.50. function FANN_GAUSSIAN
Neurons 10. Current error: 0.182559. Total error: 35.0513. Epochs 2157. Bit fail 127. candidate steepness 1.00. function FANN_GAUSSIAN_SYMMETRIC
Neurons 11. Current error: 0.171901. Total error: 33.0049. Epochs 2404. Bit fail 84. candidate steepness 0.50. function FANN_SIGMOID_SYMMETRIC
Neurons 12. Current error: 0.163253. Total error: 31.3447. Epochs 2649. Bit fail 93. candidate steepness 1.00. function FANN_GAUSSIAN_SYMMETRIC
Neurons 13. Current error: 0.160835. Total error: 30.8803. Epochs 2877. Bit fail 86. candidate steepness 0.50. function FANN_SIGMOID
Neurons 14. Current error: 0.151457. Total error: 29.0798. Epochs 3090. Bit fail 77. candidate steepness 1.00. function FANN_GAUSSIAN
Neurons 15. Current error: 0.136264. Total error: 26.1626. Epochs 3338. Bit fail 69. candidate steepness 1.00. function FANN_GAUSSIAN_SYMMETRIC
Neurons 16. Current error: 0.127493. Total error: 24.4787. Epochs 3561. Bit fail 60. candidate steepness 1.00. function FANN_GAUSSIAN_SYMMETRIC
Neurons 17. Current error: 0.120770. Total error: 23.1879. Epochs 3829. Bit fail 54. candidate steepness 1.00. function FANN_SIGMOID
Neurons 18. Current error: 0.113881. Total error: 21.8651. Epochs 4029. Bit fail 60. candidate steepness 1.00. function FANN_GAUSSIAN_SYMMETRIC
Neurons 19. Current error: 0.102948. Total error: 19.7661. Epochs 4259. Bit fail 51. candidate steepness 0.75. function FANN_GAUSSIAN_SYMMETRIC
Neurons 20. Current error: 0.084112. Total error: 16.1496. Epochs 4556. Bit fail 40. candidate steepness 1.00. function FANN_GAUSSIAN
Neurons 21. Current error: 0.076130. Total error: 14.6169. Epochs 4832. Bit fail 36. candidate steepness 1.00. function FANN_SIGMOID
Neurons 22. Current error: 0.068355. Total error: 13.1241. Epochs 5065. Bit fail 30. candidate steepness 0.75. function FANN_GAUSSIAN
Neurons 23. Current error: 0.058445. Total error: 11.2215. Epochs 5310. Bit fail 23. candidate steepness 1.00. function FANN_SIGMOID
Neurons 24. Current error: 0.046348. Total error: 8.8988. Epochs 5599. Bit fail 21. candidate steepness 0.75. function FANN_SIGMOID
Neurons 25. Current error: 0.035114. Total error: 6.7418. Epochs 5899. Bit fail 15. candidate steepness 1.00. function FANN_SIGMOID
Neurons 26. Current error: 0.027708. Total error: 5.3199. Epochs 6199. Bit fail 12. candidate steepness 1.00. function FANN_SIGMOID
Neurons 27. Current error: 0.017783. Total error: 3.4143. Epochs 6471. Bit fail 8. candidate steepness 1.00. function FANN_GAUSSIAN
Neurons 28. Current error: 0.013958. Total error: 2.6800. Epochs 6771. Bit fail 6. candidate steepness 1.00. function FANN_SIGMOID
Neurons 29. Current error: 0.006293. Total error: 1.2082. Epochs 7071. Bit fail 3. candidate steepness 1.00. function FANN_SIGMOID
Neurons 30. Current error: 0.002894. Total error: 0.5557. Epochs 7371. Bit fail 1. candidate steepness 1.00. function FANN_GAUSSIAN
Neurons 31. Current error: 0.000964. Total error: 0.1851. Epochs 7559. Bit fail 0. candidate steepness 1.00. function FANN_SIGMOID
Train outputs Current error: 0.000022. Epochs 7709