Class1 Assignment3

吴恩达DeepLearning课程作业，将其改成tensorflow版本，也是tensorflow的学习过程，记录一下。作业内容如下：

带一个隐藏层的planar data分类

from planar_utils import load_planar_dataset # 自带的加载数据方法
import tensorflow as tf

# 前向传播过程
def forward_propagation(input_tensor, W1, b1, W2, b2):
    layer1 = tf.nn.tanh(tf.matmul(input_tensor, W1) + b1)
    layer2 = tf.nn.sigmoid(tf.matmul(layer1, W2) + b2)
    return layer2

def evaluate(labels, predictions):
    ones = tf.ones_like(predictions)
    zeros = tf.zeros_like(predictions)
    predictions = tf.where(predictions < 0.5, x=zeros, y=ones)
    labels = tf.cast(labels, tf.float32)
    acc = tf.reduce_mean(tf.cast(tf.equal(labels, predictions), tf.float32))
    return acc

labels为正确答案，值为0或者1；predictions为预测的结果，值在0-1之间，因此需要转为0或1再比较，小于0.5为0，大于0.5为1

#模型训练过程
def train(X, Y, n_h, num_iteration=10000, learning_rate=1.2):
    n_x = X.shape[1]
    n_y = Y.shape[1]
    x = tf.placeholder(tf.float32, [None, n_x], name='x-input')
    y_ = tf.placeholder(tf.float32, [None, n_y], name='y-input')
    W1 =tf.Variable(tf.random_normal([n_x, n_h], stddev=0.01, seed=2))
    b1 = tf.Variable(tf.zeros([n_h]))
    W2 = tf.Variable(tf.random_normal([n_h, n_y], stddev=0.01, seed=2))
    b2 = tf.Variable(tf.zeros([n_y]))

    y = forward_propagation(x, W1, b1, W2, b2)
    loss = - tf.reduce_mean(tf.log(tf.clip_by_value(y, 1e-10, 1.0)) * y_ + tf.log(tf.clip_by_value((1-y), 1e-10, 1.0)) * (1 - y_))
    train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss)

    train_acc = evaluate(Y, y)
    with tf.Session() as sess:
        tf.global_variables_initializer().run()
        for i in range(num_iteration):
            sess.run(train_step, feed_dict={x: X, y_:Y})
            if i % 1000 == 0:
                total_loss = sess.run(loss, feed_dict={x:X, y_:Y})
                w1 = sess.run(W1)
                w2 = sess.run(W2)
                total_b1 = sess.run(b1)
                total_b2 = sess.run(b2)
                acc = sess.run(train_acc, feed_dict={x:X, y_:Y})
                print('Cost after interation %i: %f, accuracy is %f' % (i, total_loss, acc))

def main():
    X, Y = load_planar_dataset()
    train(X.T, Y.T, 4, num_iteration=20000)