简单的CNN

stock/cnn_train.py

@@ -2,17 +2,22 @@ import keras
 # -*- encoding:utf-8 -*-
 import numpy as np
 from keras.models import Sequential
-from keras.layers import Dense,Dropout
+# 优化方法选用Adam(其实可选项有很多，如SGD)
+from keras.optimizers import Adam
 import random
 from keras.models import load_model
 from imblearn.over_sampling import RandomOverSampler
 from keras.utils import np_utils
+# 用于模型初始化，Conv2D模型初始化、Activation激活函数，MaxPooling2D是池化层
+# Flatten作用是将多位输入进行一维化
+# Dense是全连接层
+from keras.layers import Conv2D, Activation, MaxPool2D, Flatten, Dense
 
 
 def read_data(path):
     lines = []
     with open(path) as f:
-        for x in range(40000):
+        for x in range(10000):
             lines.append(eval(f.readline().strip()))
 
     random.shuffle(lines)
@@ -35,19 +40,55 @@ def read_data(path):
     return X_resampled,y_resampled,np.array(test_x),np.array(test_y)
 
 train_x,train_y,test_x,test_y=read_data("D:\\data\\quantization\\stock6.log")
+train_x = train_x.reshape(train_x.shape[0], 1,80,5)
+test_x = test_x.reshape(test_x.shape[0], 1,80,5)
+# train_y = train_y.reshape(train_y.shape[0],3)
+# 调用此次训练的数据集
+from keras.datasets import mnist
+import numpy as np
+np.random.seed(1337)
+
+# 加载数据
+# (X_train,Y_train),(X_test,Y_test) = mnist.load_data()
 
 model = Sequential()
-model.add(Dense(units=635, input_dim=400,  activation='relu'))
-model.add(Dense(units=525, activation='relu'))
-model.add(Dense(units=525, activation='relu'))
-model.add(Dropout(0.2))
-model.add(Dense(units=525, activation='relu'))
-model.add(Dropout(0.2))
-model.add(Dense(units=425, activation='relu'))
-model.add(Dropout(0.2))
-model.add(Dense(units=366, activation='relu'))
-model.add(Dense(units=3, activation='softmax'))
-model.compile(loss='categorical_crossentropy', optimizer="adam",metrics=['accuracy'])
+# model.add(Dense(units=635, input_dim=400,  activation='relu'))
+# model.add(Dense(units=3, activation='softmax'))
+# model.compile(loss='categorical_crossentropy', optimizer="adam",metrics=['accuracy'])
+
+# 模型卷积层设计
+model.add(Conv2D(
+    nb_filter=32,  # 第一层设置32个滤波器
+    nb_row=3,
+    nb_col=15,  # 设置滤波器的大小为5*5
+    padding='same',  # 选择滤波器的扫描方式，即是否考虑边缘
+    input_shape=(1,80,5),  # 设置输入的形状
+    # batch_input_shape=(64, 1, 28, 28),
+))
+# 选择激活函数
+model.add(Activation('relu'))
+
+# 设置下采样(池化层）
+model.add(MaxPool2D(
+    pool_size=(2,2),  # 下采样格为2*2
+    strides=(2,2),  # 向右向下的步长
+    padding='same', # padding mode is 'same'
+))
+
+# 使用Flatten函数，将输入数据扁平化（因为输入数据是一个多维的形式，需要将其扁平化）
+model.add(Flatten())  # 将多维的输入一维化
+model.add(Dense(1024))  # 全连接层1024个点
+model.add(Activation('relu'))
+
+# 在建设一层
+model.add(Dense(3))  # 输入是个类别
+model.add(Activation('softmax'))  # 用于分类的softmax函数
+
+adam = Adam()  # 学习速率lr=0.0001
+
+model.compile(optimizer=adam,
+    loss='categorical_crossentropy',
+    metrics=['accuracy'])
 
 print("Starting training ")
 h=model.fit(train_x, train_y, batch_size=64, epochs=14, shuffle=True)