In [25]:

    

import tensorflow as tf


    

In [26]:

    

import numpy as np


    

In [27]:

    

n_inputs = 3


    

In [28]:

    

n_neurons = 5


    

In [29]:

    

X0 = tf.placeholder(tf.float32, [None, n_inputs])


    

In [30]:

    

X1 = tf.placeholder(tf.float32, [None, n_inputs])


    

In [31]:

    

Wx = tf.Variable(tf.random_normal(shape=[n_inputs, n_neurons], dtype=tf.float32))


    

In [32]:

    

Wy = tf.Variable(tf.random_normal(shape=[n_neurons, n_neurons], dtype=tf.float32))


    

In [33]:

    

b = tf.Variable(tf.zeros([1, n_neurons], dtype=tf.float32))


    

In [34]:

    

Y0 = tf.tanh(tf.matmul(X0, Wx) + b)


    

In [35]:

    

Y1 = tf.tanh(tf.matmul(X1, Wx) + tf.matmul(Y0, Wy) + b)


    

In [36]:

    

init = tf.global_variables_initializer()


    

In [37]:

    

''' This network looks much like a tow-layer feedforward neural network, with a few twists:
1. the same weights and bias terms are shared by both layers.
2. feed inputs at each layer, and get outputs from each layer.
'''


    

    
Out[37]:





' This network looks much like a tow-layer feedforward neural network, with a few twists:\n1. the same weights and bias terms are shared by both layers.\n2. feed inputs at each layer, and get outputs from each layer.\n'

In [38]:

    

X0_batch = np.array([[0, 1, 2], [3, 4, 5], [6,7, 8],[9,0,1]]) # t = 0


    

In [39]:

    

X1_batch = np.array([[9,8,7], [0,0,0],[6,5,4],[3,2,1]]) # t = 1


    

In [40]:

    

with tf.Session() as sess:
    sess.run(init)
    Y0_val, Y1_val = sess.run([Y0, Y1], feed_dict = {X0:X0_batch, X1:X1_batch})


    

In [41]:

    

print(Y0_val)


    

    




[[ 0.95313758 -0.17996429  0.25595349  0.99995422  0.95708007]
 [ 1.         -0.99662399  0.87401539  1.          1.        ]
 [ 1.         -0.99999183  0.98485768  1.          1.        ]
 [ 1.         -1.          0.99977297  1.          1.        ]]

In [42]:

    

print(Y1_val)


    

    




[[ 1.         -1.          0.99913961  1.          1.        ]
 [ 0.99348378 -0.92470729  0.59619802 -0.950019    0.96663409]
 [ 1.         -0.99999946  0.99068093  1.          1.        ]
 [ 1.         -0.99979377  0.92245167  0.99699968  1.        ]]

In [44]:

    

X3 = tf.placeholder(tf.float32, [None, n_inputs])


    

In [46]:

    

X4 = tf.placeholder(tf.float32, [None, n_inputs])


    

In [68]:

    

basic_cell = tf.nn.rnn_cell.BasicRNNCell(num_units = n_neurons)


    

In [48]:

    

output_seqs, states = tf.nn.rnn(basic_cell, [X3, X4], dtype=tf.float32)


    

In [51]:

    

Y3, Y4 = output_seqs


    

In [52]:

    

init1 = tf.global_variables_initializer()


    

In [53]:

    

with tf.Session() as sess:
    sess.run(init1)
    Y3_val, Y4_val = sess.run([Y3, Y4], feed_dict={X3:X0_batch, X4:X1_batch})


    

In [54]:

    

print(Y3_val)


    

    




[[ 0.36895558 -0.7123453  -0.91218674 -0.65849072  0.1567992 ]
 [ 0.50603181 -0.99882776 -0.99994904  0.30092672  0.50690198]
 [ 0.62156337 -0.99999589 -1.          0.88775307  0.74383038]
 [-0.45230895 -0.99997234 -0.99736899  0.99970359  0.43973276]]

In [55]:

    

print(Y4_val)


    

    




[[ 0.04022945 -0.99999976 -1.          0.99851537  0.73603499]
 [-0.58808035 -0.47645965  0.14024077 -0.30263644 -0.49604744]
 [-0.7768271  -0.99993831 -0.99998355  0.99298751 -0.1583083 ]
 [-0.80477601 -0.96781379 -0.9651143   0.92127007 -0.50737476]]

In [57]:

    

n_steps = 5


    

In [58]:

    

X = tf.placeholder(tf.float32, [None, n_steps, n_inputs])


    

In [59]:

    

X_seqs = tf.unpack(tf.transpose(X, perm=[1, 0, 2]))


    

In [66]:

    

X_seqs


    

    
Out[66]:





[<tf.Tensor 'unpack:0' shape=(?, 3) dtype=float32>,
 <tf.Tensor 'unpack:1' shape=(?, 3) dtype=float32>,
 <tf.Tensor 'unpack:2' shape=(?, 3) dtype=float32>,
 <tf.Tensor 'unpack:3' shape=(?, 3) dtype=float32>,
 <tf.Tensor 'unpack:4' shape=(?, 3) dtype=float32>]