循环神经网络(RNN)具有记忆功能,适合解决连续序列的问题,善于从具有一定顺序意义的样本与样本间学习规律。
实例:使用RNN网络拟合回声信号序列,使用一串随机的模拟数据作为原始信号,让RNN网络来拟合其对应的回声信号。
样本数据为一串随机的由0、1组成的数字,将其当成发射出去的一串信号。当碰到阻挡被反弹回来时,会收到原始信号的回音。
如果步长为3,那么输入和输出的序列如图:
import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt
num_epochs = 5
total_series_length = 50000
truncated_backprop_length = 15
state_size = 4 # RNN 状态
num_classes = 2
echo_step = 3
batch_size = 5
num_batches = total_series_length//batch_size//truncated_backprop_length
样本生成:取50000个序列样本数据,每个测试数据截取15个序列,回声序列的步长为3,最小批次为5。
# 生成样本
def generateData():
x = np.array(np.random.choice(2, total_series_length, p=[0.5, 0.5]))
y = np.roll(x, echo_step) # 向右循环移位【1111000】---【0001111】
y[0: echo_step] = 0 # 赋值为0
x = x.reshape((batch_size, -1)) # (5, 10000)
y = y.reshape((batch_size, -1))
return (x, y)
定义占位符
batchX_placeholder = tf.placeholder(tf.float32, [batch_size, truncated_backprop_length])
batchY_placeholder = tf.placeholder(tf.int32, [batch_size, truncated_backprop_length])
init_state = tf.placeholder(tf.float32, [batch_size, state_size])
inputs_series = tf.unstack(batchX_placeholder, axis=1) #
labels_series = tf.unstack(batchY_placeholder, axis=1)
定义网络
current_state = init_state
predictions_series = []
losses = []
for current_input, labels in zip(inputs_series, labels_series):
current_input = tf.reshape(current_input, [batch_size, 1])
input_and_state_concatenated = tf.concat([current_input, current_state], 1)
next_state = tf.contrib.layers.fully_connected(input_and_state_concatenated, state_size, activation_fn=tf.tanh)
current_state = next_state
logits = tf.contrib.layers.fully_connected(next_state, num_classes, activation_fn=None)
loss = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=labels, logits=logits)
losses.append(loss)
predictions = tf.nn.softmax(logits)
predictions_series.append(predictions)
total_loss = tf.reduce_mean(losses)
train_step = tf.train.AdagradOptimizer(0.3).minimize(total_loss)
训练模型与可视化
def plot(loss_list, predictions_series, batchX, batchY):
plt.subplot(2, 3, 1)
plt.cla()
plt.plot(loss_list)
for batch_series_idx in range(batch_size):
one_hot_output_series = np.array(predictions_series)[:, batch_series_idx, :]
single_output_series = np.array([(1 if out[0] < 0.5 else 0) for out in one_hot_output_series])
plt.subplot(2, 3, batch_series_idx + 2)
plt.cla()
plt.axis([0, truncated_backprop_length, 0, 2])
left_offset = range(truncated_backprop_length)
left_offset2 = range(echo_step,truncated_backprop_length+echo_step)
label1 = "past values"
label2 = "True echo values"
label3 = "Predictions"
plt.plot(left_offset2, batchX[batch_series_idx, :]*0.2+1.5, "o--b", label=label1)
plt.plot(left_offset, batchY[batch_series_idx, :]*0.2+0.8,"x--b", label=label2)
plt.plot(left_offset, single_output_series*0.2+0.1 , "o--y", label=label3)
plt.legend(loc='best')
plt.draw()
plt.pause(0.0001)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
plt.ion()
plt.figure()
plt.show()
loss_list = []
for epoch_idx in range(num_epochs):
x,y = generateData()
_current_state = np.zeros((batch_size, state_size))
print("New data, epoch", epoch_idx)
for batch_idx in range(num_batches):#50000/ 5 /15=分成多少段
start_idx = batch_idx * truncated_backprop_length
end_idx = start_idx + truncated_backprop_length
batchX = x[:,start_idx:end_idx]
batchY = y[:,start_idx:end_idx]
_total_loss, _train_step, _current_state, _predictions_series = sess.run(
[total_loss, train_step, current_state, predictions_series],
feed_dict={
batchX_placeholder:batchX,
batchY_placeholder:batchY,
init_state:_current_state
})
loss_list.append(_total_loss)
if batch_idx%100 == 0:
print("Step",batch_idx, "Loss", _total_loss)
plot(loss_list, _predictions_series, batchX, batchY)
plt.ioff()
plt.show()