training.py 文件源码

def trainMLPWithData(self, input_vector, label_vector, printSteps = 250):
    percent_split = 0.7
    trX, teX, trY, teY = cross_validation.train_test_split(input_vector, 
              label_vector, test_size=(1.0-percent_split), random_state=0)

    n_inputs = 10
    n_outputs = 8

    X = tf.placeholder("float", [None, n_inputs])
    Y = tf.placeholder("float", [None, n_outputs])

    w_h = tf.Variable(tf.random_normal([n_inputs, 10], stddev=0.01))
    w_o = tf.Variable(tf.random_normal([10, n_outputs], stddev=0.01))

    p_keep_input = tf.placeholder("float")
    p_keep_hidden = tf.placeholder("float")
    X = tf.nn.dropout(X, p_keep_input)
    h = tf.nn.relu(tf.matmul(X, w_h))
    h = tf.nn.dropout(h, p_keep_hidden)
    py_x = tf.matmul(h, w_o)

    learnRate = 0.01
    cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(py_x, Y))
    train_step = tf.train.RMSPropOptimizer(0.001, 0.9).minimize(cost)
    #train_step = tf.train.GradientDescentOptimizer(learnRate).minimize(cost)

    # Add accuracy checking nodes
    tf_correct_prediction = tf.equal(tf.argmax(py_x,1), tf.argmax(teY,1))
    tf_accuracy = tf.reduce_mean(tf.cast(tf_correct_prediction, "float"))

    # Init variables
    init = tf.initialize_all_variables()

    sess = tf.Session()
    sess.run(init)

    k=[]
    for i in range(10000):
        sess.run(train_step, feed_dict={X: trX, Y: trY, p_keep_input: 0.8, p_keep_hidden: 0.5})
        result = sess.run(tf_accuracy, feed_dict={X: teX, Y: teY, p_keep_input: 1.0, p_keep_hidden: 1.0})
        # Save data
        k.append(result)
        if (i % printSteps == 0):
          print("Run {},{}".format(i,result))

    k=np.array(k)
    print("Max accuracy: {}".format(k.max()))
    print(('MLP training with %s datapoints :: Done \n\n') % (len(input_vector)))

    self.trainedModel = sess
    return (self.trainedModel, k.max())