network.py

   1 '''
   2 Created on 2017/12/25
   3
   4 @author: fukemasashi
   5 '''
   6 from keras.models import Sequential
   7 from keras.layers import Dense,Dropout,Activation
   8 import numpy as np
   9 import os
  10
  11 class  Comp():
  12     def __init__(self):
  13         self.model1,self.model2 = Sequential(),Sequential()
  14
  15         self.model1.add(Dense(100,input_shape=(64,)))
  16         self.model1.add(Activation('relu'))
  17         self.model1.add(Dropout(0.25))
  18
  19         self.model1.add(Dense(100))
  20         self.model1.add(Activation('relu'))
  21         self.model1.add(Dropout(0.25))
  22
  23         self.model1.add(Dense(64))
  24         self.model1.add(Activation('softmax'))
  25
  26         self.model1.compile(
  27             loss='categorical_crossentropy',
  28             optimizer='adam',
  29             metrics=['accuracy'])
  30
  31         self.model2.add(Dense(100,input_shape=(64,)))
  32         self.model2.add(Activation('sigmoid'))
  33         self.model2.add(Dropout(0.25))
  34         self.model2.add(Dense(100))
  35         self.model2.add(Activation('sigmoid'))
  36         self.model2.add(Dropout(0.25))
  37         self.model2.add(Dense(64))
  38         self.model2.add(Activation('softmax'))
  39         self.model2.compile(
  40             loss='categorical_crossentropy',
  41             optimizer='adam',
  42             metrics=['accuracy'])
  43
  44     def sente_stone(self,X_train,Y_train):
  45         hdf5_file = 'sente-model.hdf5'
  46         if os.path.exists(hdf5_file):
  47             self.model1.load_weights(hdf5_file)
  48         X,Y = np.array(X_train),np.array(Y_train)
  49         X = np.reshape(np.float32(X),(1,64))
  50         Y = np.reshape(np.float32(Y),(1,64))
  51         self.model1.fit(X,Y)
  52         res = self.model1.predict(X,None,0)
  53         while True:
  54             s = np.argmax(res)
  55             if res[0][s] == 0:
  56                 s = np.argmax(Y)
  57                 print('miss')
  58             elif Y[0][s] == 0:
  59                 res[0][s] = 0
  60                 continue
  61             break
  62         self.model1.save_weights(hdf5_file)
  63         return [s // 8, s % 8]
  64
  65     def gote_stone(self,X_train,Y_train):
  66         hdf5_file = 'gote-model.hdf5'
  67         if os.path.exists(hdf5_file):
  68             self.model2.load_weights(hdf5_file)
  69         X,Y=np.array(X_train),np.array(Y_train)
  70         X = np.reshape(np.float32(X),(1,64))
  71         Y = np.reshape(np.float32(Y),(1,64))
  72         self.model2.fit(X,Y)
  73         res = self.model2.predict(X,None,0)
  74         while True:
  75             s = np.argmax(res)
  76             if res[0][s] == 0:
  77                 s = np.argmax(Y)
  78                 print('miss')
  79             elif Y[0][s] == 0:
  80                 res[0][s] = 0
  81                 continue
  82             break
  83         self.model2.save_weights(hdf5_file)
  84         return [s // 8, s % 8]