MLP_GA

多層マルチーパーセプトロンのハイパーパラメータを遺伝的アルゴリズムを用いて最適化

result exampke

....
........ 省略

 -- 9 世代 --
mate
mate
optimizer is Adam
 load mnist data
  build mlp model

        optimizer:	<keras.optimizers.Adam object at 0x7fa3475960b8>
        dense1:	256
        dense2:	32
        drop1:	0.29334818588837985
        drop2:	0.31627557353342245
        activation:	sigmoid
        batch_size:\32
        
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense_87 (Dense)             (None, 256)               200960    
_________________________________________________________________
activation_85 (Activation)   (None, 256)               0         
_________________________________________________________________
dropout_58 (Dropout)         (None, 256)               0         
_________________________________________________________________
dense_88 (Dense)             (None, 32)                8224      
_________________________________________________________________
activation_86 (Activation)   (None, 32)                0         
_________________________________________________________________
dropout_59 (Dropout)         (None, 32)                0         
_________________________________________________________________
dense_89 (Dense)             (None, 10)                330       
_________________________________________________________________
activation_87 (Activation)   (None, 10)                0         
=================================================================
Total params: 209,514
Trainable params: 209,514
Non-trainable params: 0

....
........ 省略

 4 の個体を評価 
  Min 0.07774280903284671
  Max 0.09001250675814226
  Avg 0.08313211445707129
  Std 0.004174518287204314
-- 進化終了 -- 
 最も優れていた個体 [256, 32, 0.29334818588837985, 0.31627557353342245, 32, 'sigmoid', 'Adam'] (0.07774280903284671,) 

code

各コードについて

• mlp_ga.py
  • GA(遺伝的アルゴリズム)を用いてMLPの最適モデルを導出する
• mlp.py
  • MLPクラス model.pyから呼び出す
• conv3d_ga.py
  • GAを使って3D-ConvNetの最適モデルを導出する
• conv3d.py
  • conv3DNetクラス
• OneMax_GA.py
  • GAのサンプルコード:OneMax問題をGAで解くスクリプト

Usage

1. please clone my repositori

git clone [email protected]:fuchami/MLP_GA.git

2. run script on Python3

python3 mlp_ga.py

example model

Requirement

• Software
  • python3.6.3
  • tensorflow==1.7.0
  • keras==2.1.5
  • numpy==1.14.0
  • matplotlib==2.2.2
  • deap==1.2.2