mix_train_190.py 6.2 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191
  1. import keras
  2. # -*- encoding:utf-8 -*-
  3. import numpy as np
  4. from keras.models import Sequential
  5. # 优化方法选用Adam(其实可选项有很多,如SGD)
  6. from keras.optimizers import Adam
  7. import random
  8. from keras.models import load_model
  9. from imblearn.over_sampling import RandomOverSampler
  10. from keras.utils import np_utils
  11. # 用于模型初始化,Conv2D模型初始化、Activation激活函数,MaxPooling2D是池化层
  12. # Flatten作用是将多位输入进行一维化
  13. # Dense是全连接层
  14. from keras.layers import Conv2D, Activation, MaxPool2D, Flatten, Dense,Dropout,Input,MaxPooling2D,BatchNormalization,concatenate
  15. from keras import regularizers
  16. from keras.models import Model
  17. from keras.callbacks import EarlyStopping
  18. early_stopping = EarlyStopping(monitor='accuracy', patience=5, verbose=2)
  19. epochs= 88
  20. size = 400000 #18W 60W
  21. file_path = 'D:\\data\\quantization\\stock196_18d_train2.log'
  22. model_path = '196_18d_mix_6D_ma5_s_seq.h5'
  23. file_path1='D:\\data\\quantization\\stock196_18d_test.log'
  24. '''
  25. 结果均用使用ma
  26. 4 ROC
  27. '''
  28. def read_data(path, path1=file_path1):
  29. lines = []
  30. with open(path) as f:
  31. for x in range(size): #610000
  32. line = eval(f.readline().strip())
  33. lines.append(line)
  34. # with open(path1) as f:
  35. # for x in range(50000):
  36. # line = eval(f.readline().strip())
  37. # lines.append(line)
  38. random.shuffle(lines)
  39. print('读取数据完毕')
  40. d=int(0.85*len(lines))
  41. length = len(lines[0])
  42. train_x=[s[:length - 2] for s in lines[0:d]]
  43. train_y=[s[-1] for s in lines[0:d]]
  44. test_x=[s[:length - 2] for s in lines[d:]]
  45. test_y=[s[-1] for s in lines[d:]]
  46. print('转换数据完毕')
  47. ros = RandomOverSampler(random_state=0)
  48. X_resampled, y_resampled = ros.fit_sample(np.array(train_x), np.array(train_y))
  49. print('数据重采样完毕')
  50. return X_resampled,y_resampled,np.array(test_x),np.array(test_y)
  51. train_x,train_y,test_x,test_y=read_data(file_path)
  52. train_x_a = train_x[:,:18*18]
  53. train_x_a = train_x_a.reshape(train_x.shape[0], 18, 18, 1)
  54. # train_x_b = train_x[:, 9*26:18*26]
  55. # train_x_b = train_x_b.reshape(train_x.shape[0], 9, 26, 1)
  56. train_x_c = train_x[:,18*18:]
  57. def create_mlp(dim, regress=False):
  58. # define our MLP network
  59. model = Sequential()
  60. model.add(Dense(96, input_dim=dim, activation="relu"))
  61. model.add(Dense(96, activation="relu"))
  62. # check to see if the regression node should be added
  63. if regress:
  64. model.add(Dense(1, activation="linear"))
  65. # return our model
  66. return model
  67. def create_cnn(width, height, depth, size=48, kernel_size=(5, 6), regress=False, output=24):
  68. # initialize the input shape and channel dimension, assuming
  69. # TensorFlow/channels-last ordering
  70. inputShape = (width, height, 1)
  71. chanDim = -1
  72. # define the model input
  73. inputs = Input(shape=inputShape)
  74. # x = inputs
  75. # CONV => RELU => BN => POOL
  76. x = Conv2D(size, kernel_size, strides=2, padding="same")(inputs)
  77. x = Activation("relu")(x)
  78. x = BatchNormalization(axis=chanDim)(x)
  79. # y = Conv2D(24, (2, 8), strides=2, padding="same")(inputs)
  80. # y = Activation("relu")(y)
  81. # y = BatchNormalization(axis=chanDim)(y)
  82. # flatten the volume, then FC => RELU => BN => DROPOUT
  83. x = Flatten()(x)
  84. x = Dense(output)(x)
  85. x = Activation("relu")(x)
  86. x = BatchNormalization(axis=chanDim)(x)
  87. x = Dropout(0.2)(x)
  88. # apply another FC layer, this one to match the number of nodes
  89. # coming out of the MLP
  90. x = Dense(output)(x)
  91. x = Activation("relu")(x)
  92. # check to see if the regression node should be added
  93. if regress:
  94. x = Dense(1, activation="linear")(x)
  95. # construct the CNN
  96. model = Model(inputs, x)
  97. # return the CNN
  98. return model
  99. # create the MLP and CNN models
  100. mlp = create_mlp(train_x_c.shape[1], regress=False)
  101. # cnn_0 = create_cnn(18, 21, 1, kernel_size=(3, 3), size=64, regress=False, output=128) # 31 97 46
  102. # cnn_0 = create_cnn(18, 21, 1, kernel_size=(6, 6), size=64, regress=False, output=128) # 29 98 47
  103. # cnn_0 = create_cnn(18, 21, 1, kernel_size=(9, 9), size=64, regress=False, output=128) # 28 97 53
  104. cnn_0 = create_cnn(18, 18, 1, kernel_size=(3, 18), size=96, regress=False, output=96) #A23 99 33 A' 26 99 36 #B 34 98 43
  105. # cnn_1 = create_cnn(18, 21, 1, kernel_size=(18, 11), size=96, regress=False, output=96)
  106. # cnn_1 = create_cnn(9, 26, 1, kernel_size=(2, 14), size=36, regress=False, output=64)
  107. # create the input to our final set of layers as the *output* of both
  108. # the MLP and CNN
  109. combinedInput = concatenate([mlp.output, cnn_0.output])
  110. # our final FC layer head will have two dense layers, the final one
  111. # being our regression head
  112. x = Dense(1024, activation="relu", kernel_regularizer=regularizers.l1(0.003))(combinedInput)
  113. x = Dropout(0.2)(x)
  114. x = Dense(1024, activation="relu")(x)
  115. x = Dense(1024, activation="relu")(x)
  116. # 在建设一层
  117. x = Dense(4, activation="softmax")(x)
  118. # our final model will accept categorical/numerical data on the MLP
  119. # input and images on the CNN input, outputting a single value (the
  120. # predicted price of the house)
  121. model = Model(inputs=[mlp.input, cnn_0.input,], outputs=x)
  122. print("Starting training ")
  123. # h = model.fit(train_x, train_y, batch_size=4096*2, epochs=500, shuffle=True)
  124. # compile the model using mean absolute percentage error as our loss,
  125. # implying that we seek to minimize the absolute percentage difference
  126. # between our price *predictions* and the *actual prices*
  127. opt = Adam(lr=1e-3, decay=1e-3 / 200)
  128. model.compile(loss="categorical_crossentropy", optimizer=opt, metrics=['accuracy'])
  129. # train the model
  130. print("[INFO] training model...")
  131. model.fit(
  132. [train_x_c, train_x_a,], train_y,
  133. # validation_data=([testAttrX, testImagesX], testY),
  134. # epochs=int(3*train_x_a.shape[0]/1300),
  135. epochs=epochs,
  136. batch_size=2048, shuffle=True,
  137. callbacks=[early_stopping]
  138. )
  139. model.save(model_path)
  140. test_x_a = test_x[:,:18*18]
  141. test_x_a = test_x_a.reshape(test_x.shape[0], 18, 18, 1)
  142. # test_x_b = test_x[:, 9*26:9*26+9*26]
  143. # test_x_b = test_x_b.reshape(test_x.shape[0], 9, 26, 1)
  144. test_x_c = test_x[:,18*18:]
  145. # make predictions on the testing data
  146. print("[INFO] predicting house prices...")
  147. score = model.evaluate([test_x_c, test_x_a], test_y)
  148. print(score)
  149. print('Test score:', score[0])
  150. print('Test accuracy:', score[1])