TensorFlow学习笔记（一）

简单神经网络构造

import tensorflow as tf

import matplotlib.pyplot as plt

import numpy as np

tf.set_random_seed(1)

np.random.seed(1)

# fake data

x = np.linspace(-1, 1, 100)[:, np.newaxis]          # shape (100, 1)

noise = np.random.normal(0, 0.1, size=x.shape)

y = np.power(x, 2) + noise                          # shape (100, 1) + some noise

# plot data

plt.scatter(x, y)

plt.show()

tf_x = tf.placeholder(tf.float32, x.shape)     # input x

tf_y = tf.placeholder(tf.float32, y.shape)     # input y

# neural network layers

l1 = tf.layers.dense(tf_x, 10, tf.nn.relu)          # hidden layer

output = tf.layers.dense(l1, 1)                     # output layer

loss = tf.losses.mean_squared_error(tf_y, output)   # compute cost

optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.5)

train_op = optimizer.minimize(loss)

sess = tf.Session()                                 # control training and others

sess.run(tf.global_variables_initializer())         # initialize var in graph

plt.ion()   # something about plotting 打开交互模式

for step in range(100):

    # train and net output

    _, l, pred = sess.run([train_op, loss, output], {tf_x: x, tf_y: y})

    if step % 5 == 0:

        # plot and show learning process

        plt.cla()#清除活动轴

        plt.scatter(x, y)

        plt.plot(x, pred, 'r-', lw=5)

        plt.text(0.5, 0, 'Loss=%.4f' % l, fontdict={'size': 20, 'color': 'red'})

        plt.pause(0.1)

plt.ioff()#关闭交互模式用于阻塞程序，不让图片关闭

plt.show()

优化器

import tensorflow as tf

import matplotlib.pyplot as plt

import numpy as np

tf.set_random_seed(1)

np.random.seed(1)

LR = 0.01

BATCH_SIZE = 32

# fake data

x = np.linspace(-1, 1, 100)[:, np.newaxis]          # shape (100, 1)

noise = np.random.normal(0, 0.1, size=x.shape)

y = np.power(x, 2) + noise                          # shape (100, 1) + some noise

# plot dataset

plt.scatter(x, y)

plt.show()

# default network

class Net:

    def __init__(self, opt, **kwargs):

        self.x = tf.placeholder(tf.float32, [None, 1])

        self.y = tf.placeholder(tf.float32, [None, 1])

        l = tf.layers.dense(self.x, 20, tf.nn.relu)

        out = tf.layers.dense(l, 1)

        self.loss = tf.losses.mean_squared_error(self.y, out)

        self.train = opt(LR, 货币代码**kwargs).minimize(self.loss)

# different nets

net_SGD         = Net(tf.train.GradientDescentOptimizer)

net_Momentum    = Net(tf.train.MomentumOptimizer, momentum=0.9)

net_RMSprop     = Net(tf.train.RMSPropOptimizer)

net_Adam        = Net(tf.train.AdamOptimizer)

nets = [net_SGD, net_Momentum, net_RMSprop, net_Adam]

sess = tf.Session()

sess.run(tf.global_variables_initializer())

losses_his = [[], [], [], []]   # record loss

# training

for step in range(300):          # for each training step

    index = np.random.randint(0, x.shape[0], BATCH_SIZE)

    b_x = x[index]

    b_y = y[index]

    for net, l_his in zip(nets, losses_his):

        _, l = sess.run([net.train, net.loss], {net.x: b_x, net.y: b_y})

        l_his.append(l)     # loss recoder

# plot loss history

labels = ['SGD', 'Momentum', 'RMSprop', 'Adam']

for i, l_his in enumerate(losses_his):

    plt.plot(l_his, label=labels[i])

plt.legend(loc='best')

plt.xlabel('Steps')

plt.ylabel('Loss')

plt.ylim((0, 0.2))

plt.show()