import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline

samples = 1000
domain = np.array([float(i)*0.01 for i in range(samples + 1)], np.float32)
sin = np.sin(domain)

tf.reset_default_graph()

time_step = 5
features = 1
state_size = 15

x_reshaped = tf.reshape(sin[:-1], [-1, time_step, features])
y_reshaped = tf.reshape(sin[1:], [-1, time_step, features])

x = tf.unstack(x_reshaped, axis=1)
y_ = tf.unstack(y_reshaped, axis=1)

rnn_cell = tf.nn.rnn_cell.BasicRNNCell(num_units=state_size)
output, state = tf.nn.static_rnn(cell=rnn_cell, inputs=x, dtype=tf.float32)

kernel_init = tf.truncated_normal(shape=[time_step, state_size, features])
bias_init = tf.zeros(features)

output_w = tf.Variable(kernel_init)
output_b = tf.Variable(bias_init)

y = tf.matmul(output, output_w) + output_b

print(y.shape)
print(len(y_), y_[0].shape)


for t in range(time_step):
    print(y[t].shape, y_[t].shape)