V2EX = way to explore
V2EX 是一个关于分享和探索的地方
现在注册
已注册用户请  登录
LittleUqeer
V2EX  ›  TensorFlow

Tensorflow 笔记 CNN+SVC

  •  
  •   LittleUqeer · 2017-01-10 14:51:20 +08:00 · 5716 次点击
    这是一个创建于 2634 天前的主题,其中的信息可能已经有所发展或是发生改变。

    结构 5 层卷积 - 3 层全连接 使用 SVM 取代 softmax 进行预测; 计算量有点大,大家看看即可。 卷积网络结构可以参考 AlexNet

    %%time
    import numpy as np
    import matplotlib.pylab as plt
    %matplotlib inline
    import tensorflow as tf
    from sklearn.cross_validation import train_test_split
    
    ​ fac = np.load('F:/Quotes/fac16.npy').astype(np.float32)
    ret = np.load('F:/Quotes/ret16.npy').astype(np.float32)
    
    ​ train_X, test_X, train_Y, test_Y = train_test_split(fac, ret, test_size= 0.4)
    print ('训练集 /总数据集 %.3f'%(len(train_X)/len(fac)))
    
    
    # Parameters
    learning_rate = 0.001 # 学习速率,
    training_iters = 20 # 训练次数
    batch_size = 1024 # 每次计算数量 批次大小
    display_step = 10 # 显示步长
    # Network Parameters
    n_input = 40*17 # 40 天×17 多因子
    n_classes = 7 # 根据涨跌幅度分成 7 类别
    # 这里注意要使用 one-hot 格式,也就是如果分类如 3 类 -1,0,1 则需要 3 列来表达这个分类结果, 3 类是-1 0 1 然后是哪类,哪类那一行为 1 否则为 0
    dropout = 0.5# Dropout, probability to keep units
    # tensorflow 图 Graph 输入 input ,这里的占位符均为输入
    x = tf.placeholder(tf.float32, [None, n_input])
    y = tf.placeholder(tf.float32, [None, n_classes])
    keep_prob = tf.placeholder(tf.float32) #dropout (keep probability)
    
    ​# 2 层 CNN 提取特征向量
    def CNN_Net_two(x,weights,biases,dropout=0.8,m=1):
        # layer hidden 1
        x = tf.reshape(x, shape=[-1,40,17,1])
        x = tf.nn.conv2d(x, weights['wc1'], strides=[1,m,m,1],padding='SAME')
        x = tf.nn.bias_add(x,biases['bc1'])
        x = tf.nn.relu(x)
        x = tf.nn.local_response_normalization(x, depth_radius=5, bias=1.0, alpha=0.001/9.0)
        x = tf.nn.dropout(x,0.3)
    
        # layer hidden 2
        x = tf.nn.conv2d(x, weights['wc2'], strides=[1,m,m,1],padding='SAME')
        x = tf.nn.bias_add(x,biases['bc2'])
        x = tf.nn.relu(x)
        x = tf.nn.local_response_normalization(x, depth_radius=5, bias=1.0, alpha=0.001/9.0)
        x = tf.nn.dropout(x,0.3)
        
        # layer hidden 3
        x = tf.nn.conv2d(x, weights['wc3'], strides=[1,m,m,1],padding='SAME')
        x = tf.nn.bias_add(x,biases['bc3'])
        x = tf.nn.relu(x)
        x = tf.nn.local_response_normalization(x, depth_radius=5, bias=1.0, alpha=0.001/9.0)
        x = tf.nn.dropout(x,0.3)
    
        # layer hidden 4
        x = tf.nn.conv2d(x, weights['wc4'], strides=[1,m,m,1],padding='SAME')
        x = tf.nn.bias_add(x,biases['bc4'])
        x = tf.nn.relu(x)
        x = tf.nn.local_response_normalization(x, depth_radius=5, bias=1.0, alpha=0.001/9.0)
        x = tf.nn.dropout(x,0.3)
    
        # layer hidden 5
        x = tf.nn.conv2d(x, weights['wc5'], strides=[1,m,m,1],padding='SAME')
        x = tf.nn.bias_add(x,biases['bc5'])
        x = tf.nn.relu(x)
        x = tf.nn.local_response_normalization(x, depth_radius=5, bias=1.0, alpha=0.001/9.0)
        x = tf.nn.dropout(x,0.3)
        #print (x.get_shape().as_list())
    
        # 全连接层 1
        x = tf.reshape(x,[-1,weights['wd1'].get_shape().as_list()[0]])
        x = tf.add(tf.matmul(x,weights['wd1']),biases['bd1'])
        x = tf.nn.relu(x)
        x = tf.nn.dropout(x,dropout)
        #print (x.get_shape().as_list())
    
        # 全连接层 2
        x = tf.reshape(x,[-1,weights['wd2'].get_shape().as_list()[0]])
        x = tf.add(tf.matmul(x,weights['wd2']),biases['bd2'])
        x = tf.nn.relu(x)
        x = tf.nn.dropout(x,dropout)
        #print (x.get_shape().as_list())
    
        # 全连接层 3
        x = tf.reshape(x,[-1,weights['wd3'].get_shape().as_list()[0]])
        x = tf.add(tf.matmul(x,weights['wd3']),biases['bd3'])
        x = tf.nn.relu(x)
        x = tf.nn.dropout(x,dropout)
        #print (x.get_shape().as_list())
    
        
        t = tf.add(tf.matmul(x,weights['out']),biases['out'])
        #print (t.get_shape().as_list())
        # 返回两个数值, t 用于 softmax 分类, x 用于提取 CNN 处理的数据,也就是经过卷积处理的特征向量。
        return t,x
    
    ​
    # Store layers weight & bias
    weights = {
        'wc1': tf.Variable(tf.random_normal([10, 5, 1, 64])),
        'wc2': tf.Variable(tf.random_normal([10, 5, 64, 128])),
        'wc3': tf.Variable(tf.random_normal([10, 5, 128, 256])),
        'wc4': tf.Variable(tf.random_normal([10, 5, 256, 512])),
        'wc5': tf.Variable(tf.random_normal([10, 5, 512, 1024])),
        'wd1': tf.Variable(tf.random_normal([40*17*1024, 1024])),
        'wd2': tf.Variable(tf.random_normal([1024, 256])),
        'wd3': tf.Variable(tf.random_normal([256, 32])),
        'out': tf.Variable(tf.random_normal([32, n_classes]))
    }
    
    ​ biases = {
    
        'bc1': tf.Variable(tf.random_normal([64])),
        'bc2': tf.Variable(tf.random_normal([128])),
        'bc3': tf.Variable(tf.random_normal([256])),
        'bc4': tf.Variable(tf.random_normal([512])),
        'bc5': tf.Variable(tf.random_normal([1024])),
        'bd1': tf.Variable(tf.random_normal([1024])),
        'bd2': tf.Variable(tf.random_normal([256])),
        'bd3': tf.Variable(tf.random_normal([32])),
        'out': tf.Variable(tf.random_normal([n_classes]))
    }
    # 模型优化
    pred,tmp = CNN_Net_two(x,weights,biases,dropout=keep_prob)
    cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(pred,y))
    optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)
    correct_pred = tf.equal(tf.argmax(pred,1),tf.arg_max(y,1))
    # tf.argmax(input,axis=None) 由于标签的数据格式是 -1 0 1 3 列,该语句是表示返回值最大也就是 1 的索引,两个索引相同则是预测正确。
    accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))
    # 更改数据格式,降低均值
    init = tf.global_variables_initializer()
    
    

    计算保存模型

    saver = tf.train.Saver()
    with tf.Session() as sess:
        sess.run(init)
        # for step in range(300):
        for step in range(1):
            trl=int(len(train_X)/batch_size)
            for i in range(trl):
                print (i,'--',trl)
                batch_x = train_X[i*batch_size:(i+1)*batch_size]
                batch_y = train_Y[i*batch_size:(i+1)*batch_size]
                sess.run(optimizer,feed_dict={x:batch_x,y:batch_y,keep_prob:0.5})
            loss, acc = sess.run([cost, accuracy], feed_dict={x: batch_x,y: batch_y,keep_prob: 1.})
            print("Iter " + str(step*batch_size) + ", Minibatch Loss= " + \
                      "{:.6f}".format(loss) + ", Training Accuracy= " + \
                      "{:.5f}".format(acc))
        save_path = saver.save(sess,'F:/Quotes/test_var.ckpt')
        print ('保持变量')
        print("Optimization Finished!")   
        sess.close()
    
    

    读取模型,进行预测

    saver = tf.train.Saver()
    with tf.Session() as sess:
        sess.run(init)
        saver.restore(sess,'F:/Quotes/test_var.ckpt')
        trainX_Convolution = sess.run(tmp, feed_dict={x:train_X, keep_prob:1.})
        # 经过卷积处理的特征向量
        nn_score = sess.run(accuracy,feed_dict={x:train_X, keep_prob:1.})
        nn_score1 = sess.run(accuracy,feed_dict={x:test_X, keep_prob:1.})
        print(nn_score,'---',nn_score1)
        sess.close()
    
    

    one-hot 向量转换为列向量

    # train_Y 
    ol_train_Y = []
    for i in range(len(train_Y)):
        t = train_Y[i]
        arg = np.argmax(t)
        ol_train_Y.append(arg)
        
    # softmax_pred 
    ol_softmax_pred = []
    for i in range(len(softmax_pred)):
        t = softmax_pred [i]
        arg = np.argmax(t)
        ol_softmax_pred.append(arg)
    
    

    SVM 预测

    from sklearn.svm import SVC
    clf = SVC(C=0.9,gamma=1.0,decision_function_shape='ovo')
    clf.fit(trainX_Convolution, ol_train_Y)
    c = clf.predict(trainX_Convolution)
    print ('CNN 预测',(np.corrcoef(a,c)[0][1]))
    
    

    集成算法比较参见: https://uqer.io/community/share/58562a9f6a5e6d0052291ebe

    目前尚无回复
    关于   ·   帮助文档   ·   博客   ·   API   ·   FAQ   ·   我们的愿景   ·   实用小工具   ·   1124 人在线   最高记录 6543   ·     Select Language
    创意工作者们的社区
    World is powered by solitude
    VERSION: 3.9.8.5 · 33ms · UTC 22:52 · PVG 06:52 · LAX 15:52 · JFK 18:52
    Developed with CodeLauncher
    ♥ Do have faith in what you're doing.