[TensorFlow深度学习入门]实战十·用RNN(LSTM)做时间序列预测（曲线拟合）

%matplotlib inline
import os
os.environ["KMP_DUPLICATE_LIB_OK"]="TRUE"
import numpy as np 
import matplotlib.pyplot as plt
import tensorflow as tf


# hyperparameters
lr = 0.002                  # learning rate
training_iters = 500     # train step 上限
batch_size = 30            
n_inputs = 1               # MNIST data input (img shape: 28*28)
n_steps = 10                # time steps
n_hidden_units = 16        # neurons in hidden layer
n_classes = 1              # MNIST classes (0-9 digits)

def get_data(x,w,b):
    c,r = x.shape
    y = np.sin(w*x) + b + (0.01*(2*np.random.rand(c,r)-1))
    return(y)

xs = np.arange(0,3,0.01).reshape(-1,1)
ys = get_data(xs,5,0.5) 
datas = []
for i in range(len(xs)-11):
    datas.append(ys[i:i+11])
datas = np.array(datas).reshape(-1,11)
print(datas.shape)
plt.title("curve")
plt.plot(ys)
plt.show()

(289, 11)

#mnist.train.images.shape
# x y placeholder
x = tf.placeholder(tf.float32, [None, n_steps*n_inputs])
y = tf.placeholder(tf.float32, [None, n_classes])
# 对 weights biases 初始值的定义
weights = {
    # shape (16, 1)
    'out': tf.Variable(tf.random_normal([n_hidden_units, n_classes]))
}
biases = {
    # shape (1, 1)
    'out': tf.Variable(tf.constant(0.1, shape=[1, n_classes]))
}

def RNN(X, weights, biases):
    # 原始的 X  2 维数据(-1,10)
    # X ==> (-1 batches , 10 steps, 1 inputs)
    X = tf.reshape(X, [-1,n_steps,n_inputs])
    #lstm_cell (-1,10,16)
    lstm_cell = tf.nn.rnn_cell.BasicLSTMCell(n_hidden_units, forget_bias = 1.0, state_is_tuple = True)
    #print(lstm_cell)
    _init_state = lstm_cell.zero_state(289,dtype=tf.float32)
    output, states = tf.nn.dynamic_rnn(lstm_cell,X,initial_state=_init_state,time_major=False)
    print(output) #(-1,10,16)
    #finial output 
    result = tf.matmul(output[:,-1,:],weights["out"]+biases["out"])

    return(result)

pred = RNN(x, weights, biases)
cost = tf.reduce_mean(tf.square(pred-y))
train_op = tf.train.AdamOptimizer(lr).minimize(cost)


init = tf.global_variables_initializer()

with tf.Session() as sess:
    sess.run(init)
    srun = sess.run
    
    for t in range(training_iters+1):
        
        srun(train_op,{x:datas[0:289,:10],y:datas[0:289,10:11]})
        if(t%10 == 0):
            loss_val = srun(cost,{x:datas[0:289,:10],y:datas[0:289,10:11]})
            print(t,loss_val)
    y_val = srun(pred,{x:datas[0:289,:10]}).reshape(-1,1)
    plt.title("pre")
    plt.plot(y_val)
    plt.show()

WARNING:tensorflow:From <ipython-input-2-130bdeb48069>:20: BasicLSTMCell.__init__ (from tensorflow.python.ops.rnn_cell_impl) is deprecated and will be removed in a future version.
Instructions for updating:
This class is deprecated, please use tf.nn.rnn_cell.LSTMCell, which supports all the feature this cell currently has. Please replace the existing code with tf.nn.rnn_cell.LSTMCell(name='basic_lstm_cell').
Tensor("rnn/transpose_1:0", shape=(289, 10, 16), dtype=float32)
0 2.7168088
10 1.0216647
20 0.29450005
30 0.16755253
...
470 0.0010900635
480 0.001046965
490 0.001006315
500 0.0009679485