深層学習を作成する際にtf.nn.rnn_cell.BasicRNNCellクラスをtf.nn.rnn

前提・実現したいこと

お世話になっております。

1)現在、時系列データの予測を行う深層学習の作成を以下のwebサイトをもとに行っております。
URL：https://www.atmarkit.co.jp/ait/articles/1804/26/news150_2.html

2)このサイトではTensorflowのtf.nn.rnn_cell.BasicRNNCellクラスを使用して深層学習を作成しています。

3)webサイトをもとにしたプログラムを実行していく中でセルが複雑なRNN（LSTM）のモデルを作成できるtf.nn.rnn_cell.BasicLSTMCellクラスを使用してみたくなり、tf.nn.rnn_cell.BasicRNNCellクラスをを元として書き換えています。

4)しかし、tf.nn. rnn_cell.BasicLSTMCellクラスに書き換え、実行しようとした際にエラーメッセージが出力され、どのように対応したらいいかわからず手詰まりになってしまいました。

5)どのようにすればこのエラーを解消し、プログラムを実行できますでしょうか？

厚かましく申し訳ありませんが、アドバイスやご回答いただければ幸いです。
また、質問で不足している事項がありましたら、ご指摘ください。よろしくお願いします。

試したこと

1)tf.nn.rnn_cell.BasicRNNCellクラスを用いた際のcodeは以下の通りです。

python
1cell = tf.nn.rnn_cell.BasicRNNCell(10)
2initial_state = cell.zero_state(tf.shape(x)[0], dtype=tf.float32)
3outputs, last_state = tf.nn.dynamic_rnn(cell, x, initial_state=initial_state, dtype=tf.float32)

2)上記のcodeをもとにtf.nn. rnn_cell.BasicLSTMCellクラスに書き換えたcodeが以下になります。

python
1cell = tf.nn.rnn_cell.BasicLSTMCell(100)
2initial_state_c = tf.placeholder(tf.float32, [None, 100])
3initial_state_h = tf.placeholder(tf.float32, [None, 100])
4initial_state = (initial_state_c, initial_state_h)
5outputs, last_state = tf.nn.dynamic_rnn(cell, x, initial_state=initial_state, dtype=tf.float32)

3)2)のプログラムを実行した際に出力されたエラーメッセージが以下になります。

python
1---------------------------------------------------------------------------
2TypeError                                 Traceback (most recent call last)
3c:\PATH\anaconda3\envs\tensor\lib\site-packages\tensorflow_core\python\util\nest.py in assert_same_structure(nest1, nest2, check_types, expand_composites)
4    318     _pywrap_tensorflow.AssertSameStructure(nest1, nest2, check_types,
5--> 319                                            expand_composites)
6    320   except (ValueError, TypeError) as e:
7
8TypeError: The two structures don't have the same nested structure.
9
10First structure: type=list str=[<tf.Tensor 'rnn/while/Identity:0' shape=() dtype=int32>, (<tf.Tensor 'rnn/while/Identity_1:0' shape=() dtype=int32>, (<tensorflow.python.ops.tensor_array_ops.TensorArray object at 0x00000111B1B9B588>,), (<tf.Tensor 'rnn/while/Identity_3:0' shape=(?, 100) dtype=float32>, <tf.Tensor 'rnn/while/Identity_4:0' shape=(?, 100) dtype=float32>))]
11
12Second structure: type=list str=[<tf.Tensor 'rnn/while/add:0' shape=() dtype=int32>, (<tf.Tensor 'rnn/while/add_1:0' shape=() dtype=int32>, (<tensorflow.python.ops.tensor_array_ops.TensorArray object at 0x00000111B1B04B08>,), LSTMStateTuple(c=<tf.Tensor 'rnn/while/basic_lstm_cell/Add_1:0' shape=(?, 100) dtype=float32>, h=<tf.Tensor 'rnn/while/basic_lstm_cell/Mul_2:0' shape=(?, 100) dtype=float32>))]
13
14More specifically: The two namedtuples don't have the same sequence type. First structure type=tuple str=(<tf.Tensor 'rnn/while/Identity_3:0' shape=(?, 100) dtype=float32>, <tf.Tensor 'rnn/while/Identity_4:0' shape=(?, 100) dtype=float32>) has type tuple, while second structure type=LSTMStateTuple str=LSTMStateTuple(c=<tf.Tensor 'rnn/while/basic_lstm_cell/Add_1:0' shape=(?, 100) dtype=float32>, h=<tf.Tensor 'rnn/while/basic_lstm_cell/Mul_2:0' shape=(?, 100) dtype=float32>) has type LSTMStateTuple
15
16During handling of the above exception, another exception occurred:
17
18TypeError                                 Traceback (most recent call last)
19<ipython-input-10-a3291a61f68f> in <module>
20      4 initial_state_h = tf.placeholder(tf.float32, [None, 100])    #前のcellの出力を参照するのでその初期値
21      5 initial_state = (initial_state_c, initial_state_h)
22----> 6 outputs, last_state = tf.nn.dynamic_rnn(cell, x, initial_state=initial_state, dtype=tf.float32)
23
24c:\PATH\anaconda3\envs\tensor\lib\site-packages\tensorflow_core\python\util\deprecation.py in new_func(*args, **kwargs)
25    322               'in a future version' if date is None else ('after %s' % date),
26    323               instructions)
27--> 324       return func(*args, **kwargs)
28    325     return tf_decorator.make_decorator(
29    326         func, new_func, 'deprecated',
30
31c:\PATH\anaconda3\envs\tensor\lib\site-packages\tensorflow_core\python\ops\rnn.py in dynamic_rnn(cell, inputs, sequence_length, initial_state, dtype, parallel_iterations, swap_memory, time_major, scope)
32    705         swap_memory=swap_memory,
33    706         sequence_length=sequence_length,
34--> 707         dtype=dtype)
35    708 
36    709     # Outputs of _dynamic_rnn_loop are always shaped [time, batch, depth].
37
38c:\PATH\anaconda3\envs\tensor\lib\site-packages\tensorflow_core\python\ops\rnn.py in _dynamic_rnn_loop(cell, inputs, initial_state, parallel_iterations, swap_memory, sequence_length, dtype)
39    914       parallel_iterations=parallel_iterations,
40    915       maximum_iterations=time_steps,
41--> 916       swap_memory=swap_memory)
42    917 
43    918   # Unpack final output if not using output tuples.
44
45c:\PATH\anaconda3\envs\tensor\lib\site-packages\tensorflow_core\python\ops\control_flow_ops.py in while_loop(cond, body, loop_vars, shape_invariants, parallel_iterations, back_prop, swap_memory, name, maximum_iterations, return_same_structure)
46   2751       ops.add_to_collection(ops.GraphKeys.WHILE_CONTEXT, loop_context)
47   2752     result = loop_context.BuildLoop(cond, body, loop_vars, shape_invariants,
48-> 2753                                     return_same_structure)
49   2754     if maximum_iterations is not None:
50   2755       return result[1]
51
52c:\PATH\anaconda3\envs\tensor\lib\site-packages\tensorflow_core\python\ops\control_flow_ops.py in BuildLoop(self, pred, body, loop_vars, shape_invariants, return_same_structure)
53   2243       with ops.get_default_graph()._mutation_lock():  # pylint: disable=protected-access
54   2244         original_body_result, exit_vars = self._BuildLoop(
55-> 2245             pred, body, original_loop_vars, loop_vars, shape_invariants)
56   2246     finally:
57   2247       self.Exit()
58
59c:\PATH\anaconda3\envs\tensor\lib\site-packages\tensorflow_core\python\ops\control_flow_ops.py in _BuildLoop(self, pred, body, original_loop_vars, loop_vars, shape_invariants)
60   2192     # outputs of the body are typically tuples.
61   2193     nest.assert_same_structure(
62-> 2194         list(packed_vars_for_body), list(body_result), expand_composites=True)
63   2195 
64   2196     # Store body_result to keep track of TensorArrays returned by body
65
66c:\PATH\anaconda3\envs\tensor\lib\site-packages\tensorflow_core\python\util\nest.py in assert_same_structure(nest1, nest2, check_types, expand_composites)
67    324                   "Entire first structure:\n%s\n"
68    325                   "Entire second structure:\n%s"
69--> 326                   % (str(e), str1, str2))
70    327 
71    328 
72
73TypeError: The two structures don't have the same nested structure.
74
75First structure: type=list str=[<tf.Tensor 'rnn/while/Identity:0' shape=() dtype=int32>, (<tf.Tensor 'rnn/while/Identity_1:0' shape=() dtype=int32>, (<tensorflow.python.ops.tensor_array_ops.TensorArray object at 0x00000111B1B9B588>,), (<tf.Tensor 'rnn/while/Identity_3:0' shape=(?, 100) dtype=float32>, <tf.Tensor 'rnn/while/Identity_4:0' shape=(?, 100) dtype=float32>))]
76
77Second structure: type=list str=[<tf.Tensor 'rnn/while/add:0' shape=() dtype=int32>, (<tf.Tensor 'rnn/while/add_1:0' shape=() dtype=int32>, (<tensorflow.python.ops.tensor_array_ops.TensorArray object at 0x00000111B1B04B08>,), LSTMStateTuple(c=<tf.Tensor 'rnn/while/basic_lstm_cell/Add_1:0' shape=(?, 100) dtype=float32>, h=<tf.Tensor 'rnn/while/basic_lstm_cell/Mul_2:0' shape=(?, 100) dtype=float32>))]
78
79More specifically: The two namedtuples don't have the same sequence type. First structure type=tuple str=(<tf.Tensor 'rnn/while/Identity_3:0' shape=(?, 100) dtype=float32>, <tf.Tensor 'rnn/while/Identity_4:0' shape=(?, 100) dtype=float32>) has type tuple, while second structure type=LSTMStateTuple str=LSTMStateTuple(c=<tf.Tensor 'rnn/while/basic_lstm_cell/Add_1:0' shape=(?, 100) dtype=float32>, h=<tf.Tensor 'rnn/while/basic_lstm_cell/Mul_2:0' shape=(?, 100) dtype=float32>) has type LSTMStateTuple
80Entire first structure:
81[., (., (.,), (., .))]
82Entire second structure:
83[., (., (.,), LSTMStateTuple(c=., h=.))]