TensorflowのRNNを用いたsin波の予測を行いたい

前提・実現したいこと

TensorflowのRNNを使って, sin波の予測を行いたいです.
まずは簡単なsin波の予測を行い, Tensorflowについて理解を深めてから研究に応用したいと思っています.

発生している問題・エラーメッセージ

以下のサイトを参考にコーディングを行いましたが, エラーが発生してしまいます.
https://qiita.com/p-baleine@github/items/262399318c88f5b069ba

特に, 下記の部分が何をやっているのかがわからなく, この部分でエラーが発生しています.

隠れ層～出力層
outputs, state = tf.compat.v1.nn.dynamic_rnn(cell, x, sequence_length=sequence_length, initial_state=initial_state, dtype=tf.float32,
parallel_iterations=None, swap_memory=False, time_major=False, scope=None
)

index = tf.range(0, batch_size) * lenge + (sequence_length - 1)
output = tf.gather(tf.reshape(output, [-1, n_hidden]), index)

以下は, エラーメッセージをpdf化したものです.
エラーメッセージ

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-126-cba0a54bd80a> in <module>
     71 #隠れ層～出力層
     72 outputs, state = tf.compat.v1.nn.dynamic_rnn(cell, x, sequence_length=sequence_length, initial_state=initial_state, dtype=tf.float32, 
---> 73                                              parallel_iterations=None, swap_memory=False, time_major=False, scope=None
     74                                             )
     75

該当のソースコード

Python3
1import matplotlib.pyplot as plt
2import numpy as np
3import tensorflow as tf
4import random
5import time
6import os
7import csv 
8
9def init_weight_variable(shape):
10    """Initialize variable in a suitable way for weights."""
11    initial = tf.truncated_normal(shape, stddev=0.1)
12    return tf.Variable(initial)
13
14def init_bias_variable(shape):
15    """Initialize variable in a suitable way for biases."""
16    initial = tf.constant(0.1, shape=shape)
17    return tf.Variable(initial)
18    
19def make_data(data_num):
20    amp = 10
21    DATA = np.linspace(0., 2 * np.pi, data_num)
22    DATA = DATA[:, None]
23    data = amp*np.sin(DATA)
24    data = data + 0.1 * amp * ((1 - (-1)) * np.random.rand(data_num, n_input) + (-1))
25    data.reshape(data_num, n_input)
26    return data
27
28def data_to_sequence(data):
29    lenge = 25
30    TIME = data.shape[0]
31    period = TIME - lenge
32    x = []
33    y = []
34    for i in range(0, period):
35        x.append(data[i: i+lenge])
36        y.append(data[i+lenge])
37    #print(f"{y}")
38    npx = np.array(x).reshape(period, lenge, n_input)
39    npy = np.array(y).reshape(period, n_input)
40    return npx, npy
41
42
43t1 = time.time() 
44# Step 1. Prepare data
45
46#学習データ
47traindata_num = 30
48traindata = make_data(traindata_num)
49train_x, train_y = data_to_sequence(traindata)
50print(str(train_x.shape) +"," + str(train_y.shape))
51
52#テストデータ
53testdata_num = 30
54testdata = make_data(testdata_num)
55test_x, test_y = data_to_sequence(testdata)
56
57# Step 2. Define operation
58n_input = 1
59n_hidden = 30
60n_output = 1
61
62#入力層
63x = tf.placeholder(tf.float32, [None, lenge, n_input])
64sequence_length = tf.placeholder(tf.int32, [None])
65batch_size = tf.shape(x)[0]
66
67#隠れ層
68cell = tf.compat.v1.nn.rnn_cell.LSTMCell(n_hidden)
69initial_state = cell.zero_state(batch_size, dtype=tf.float32)
70
71#隠れ層～出力層
72outputs, state = tf.compat.v1.nn.dynamic_rnn(cell, x, sequence_length=sequence_length, initial_state=initial_state, dtype=tf.float32, 
73                                             parallel_iterations=None, swap_memory=False, time_major=False, scope=None
74                                            )
75
76index = tf.range(0, batch_size) * lenge + (sequence_length - 1)
77output = tf.gather(tf.reshape(output, [-1, n_hidden]), index)
78
79w = (init_weight_variable((n_hidden, n_output)))
80b = (init_bias_variable((n_output,)))
81y_output = tf.matmul((output, w) + b)
82
83init = tf.global_variables_initializer()
84y = tf.placeholder(tf.float32)
85loss = tf.reduce_mean((y_output - y)**2)
86train = tf.train.AdamOptimizer().minimize(loss)
87
88# Step 3. Run operation
89with tf.Session() as sess:
90    sess.run(init)
91    epoch = 1000
92    batch_size = 32 # バッチサイズ
93    iteration = int(train_x.size/(n_input*batch_size))
94    vec_loss = []
95    print("学習開始")
96    for i in range(epoch):
97        # ランダムなindexを作成（0〜学習データのサイズをバッチの個数分)
98        rnd_idx = np.random.randint(0, int(TIME - lenge))
99        for j in range(iteration):
100            x_batch_data = train_x[rnd_idx]
101            y_batch_data = train_y[rnd_idx]
102            sess.run(train, feed_dict={x: x_batch_data, y: y_batch_data})
103        
104        loss_train = sess.run(loss, feed_dict={x: train_x, y: train_y})
105        vec_loss.append(loss_train)
106        if i % (epoch/10) == 0 and i != 0 or i == 1:
107            print(i,"epoch, " f"Loss: {loss_train}")
108            train_y_output = sess.run(y_output, {x: train_x, y: train_y})
109            plt.plot(train_y, label='Train_answer')
110            plt.plot(train_output, label='Train_output')
111            plt.legend()
112            plt.show()
113     
114    train_output = sess.run(y_output, {x: train_x, y: train_y})
115    plt.plot(train_y, label='Train_answer')
116    plt.plot(train_output, label='Train_output')
117    plt.legend()
118    plt.show()
119      
120    plt.plot(vec_loss, label='Error of training')
121    plt.legend()
122    plt.show()
123       
124    loss_test = sess.run(loss, feed_dict={x: test_x, y: test_y})
125    print("Testdata:, " f"Loss: {loss_test}")
126    
127    plt.plot(test_y, label='Test_answer')
128    plt.plot(test_output, label='Test_output')
129    plt.legend()
130    plt.show()
131
132t2 = time.time() 
133
134# 経過時間を表示
135elapsed_time = t2-t1
136print(f"経過時間：{elapsed_time}")