RNN学习之时间序列预测sin正弦函数

在学习《Tensorflow实战Google深度学习框架》的循环神经网络应用样例：预测sin正弦函数的时间序列问题。源码运行一直有问题。如下错误：

ValueError: Trying to share variable rnn/multi_rnn_cell/cell_0/basic_lstm_cell/kernel, but specified shape (60, 120) and found shape (40, 120).

找原因，网上看到有人从多层LSTM入手，改为list添加两层Cell。

• 原Code

    lstm_cell = tf.contrib.rnn.BasicLSTMCell(HIDDEN_SIZE, state_is_tuple=True)
    cell = tf.contrib.rnn.MultiRNNCell([lstm_cell] * NUM_LAYERS)

• 修改后的Code

    lstm_model = []
    for i in range(2):
        lstm_model.append(tf.contrib.rnn.BasicLSTMCell(HIDDEN_SIZE, state_is_tuple=True))
    cell = tf.contrib.rnn.MultiRNNCell(cells=lstm_model, state_is_tuple=True)

1. 完整代码

备注：学习，抠的书籍作者的代码，见谅。

import numpy as np
import tensorflow as tf
from tensorflow.contrib.learn.python.learn.estimators.estimator import SKCompat
from tensorflow.python.ops import array_ops as array_ops_
import matplotlib.pyplot as plt
learn = tf.contrib.learn


HIDDEN_SIZE = 30
NUM_LAYERS = 2

TIMESTEPS = 10
TRAINING_STEPS = 10000
BATCH_SIZE = 32

TRAINING_EXAMPLES = 10000
TESTING_EXAMPLES = 1000
SAMPLE_GAP = 0.01


def generate_data(seq):
    """生成正弦数据"""
    X = []
    y = []

    for i in range(len(seq) - TIMESTEPS):
        X.append([seq[i: i + TIMESTEPS]])
        y.append([seq[i + TIMESTEPS]])
    return np.array(X, dtype=np.float32), np.array(y, dtype=np.float32)

def lstm_model(X, y):
    # 将如下两行代码注释掉

    # lstm_cell = tf.contrib.rnn.BasicLSTMCell(HIDDEN_SIZE, state_is_tuple=True)
    # cell = tf.contrib.rnn.MultiRNNCell([lstm_cell] * NUM_LAYERS)

    # 以上两行代码改为如下四行：
    lstm_model = []
    for i in range(2):
        lstm_model.append(tf.contrib.rnn.BasicLSTMCell(HIDDEN_SIZE, state_is_tuple=True))
    cell = tf.contrib.rnn.MultiRNNCell(cells=lstm_model, state_is_tuple=True)
    
    output, _ = tf.nn.dynamic_rnn(cell, X, dtype=tf.float32)
    output = tf.reshape(output, [-1, HIDDEN_SIZE])
    
    # 通过无激活函数的全联接层计算线性回归，并将数据压缩成一维数组的结构。
    predictions = tf.contrib.layers.fully_connected(output, 1, None)
    
    # 将predictions和labels调整统一的shape
    labels = tf.reshape(y, [-1])
    predictions=tf.reshape(predictions, [-1])
    
    loss = tf.losses.mean_squared_error(predictions, labels)
    
    train_op = tf.contrib.layers.optimize_loss(
        loss, tf.contrib.framework.get_global_step(),
        optimizer="Adagrad", learning_rate=0.1)

    return predictions, loss, train_op

if __name__ == '__main__':
    regressor = SKCompat(learn.Estimator(model_fn=lstm_model,model_dir="../src/RNN_Tensorflow/TimeSeries/Models/model_2"))

    # 生成数据。
    test_start = TRAINING_EXAMPLES * SAMPLE_GAP
    test_end = (TRAINING_EXAMPLES + TESTING_EXAMPLES) * SAMPLE_GAP
    train_X, train_y = generate_data(np.sin(np.linspace(
        0, test_start, TRAINING_EXAMPLES, dtype=np.float32)))
    print(train_X.shape)
    print(train_y.shape)
    test_X, test_y = generate_data(np.sin(np.linspace(
        test_start, test_end, TESTING_EXAMPLES, dtype=np.float32)))

    # 拟合数据。
    regressor.fit(train_X, train_y, batch_size=BATCH_SIZE, steps=TRAINING_STEPS)

    # 计算预测值。
    predicted = [[pred] for pred in regressor.predict(test_X)]

    # 计算MSE。
    rmse = np.sqrt(((predicted - test_y) ** 2).mean(axis=0))
    print ("Mean Square Error is: %f" % rmse[0])

    plot_predicted, = plt.plot(predicted, label='predicted', linestyle='-', marker='*')
    plot_test, = plt.plot(test_y, label='real_sin')
    plt.legend([plot_predicted, plot_test],['predicted', 'real_sin'])
    plt.show()

2. 效果