Ошибка сброса Tensorflow распределенной функции - PullRequest
Новая
       24

Ошибка сброса Tensorflow распределенной функции

0 голосов
/ 05 февраля 2020

Я новичок в ML и tenorflow и пытаюсь обучить и использовать стандартную модель генерации текста. Когда я go обучаю модель, я получаю эту ошибку : 

Train for 155 steps
Epoch 1/5
  2/155 [..............................] - ETA: 4:49 - loss: 2.5786
---------------------------------------------------------------------------
InvalidArgumentError                      Traceback (most recent call last)
<ipython-input-133-d70c02ff4270> in <module>()
----> 1 model.fit(dataset, epochs=epochs, callbacks=[checkpoint_callback])

11 frames
/usr/local/lib/python3.6/dist-packages/six.py in raise_from(value, from_value)

InvalidArgumentError: 2 root error(s) found.
  (0) Invalid argument:  indices[58,87] = 63 is not in [0, 63)
     [[node sequential_12/embedding_12/embedding_lookup (defined at <ipython-input-131-d70c02ff4270>:1) ]]
     [[VariableShape/_24]]
  (1) Invalid argument:  indices[58,87] = 63 is not in [0, 63)
     [[node sequential_12/embedding_12/embedding_lookup (defined at <ipython-input-131-d70c02ff4270>:1) ]]
0 successful operations.
0 derived errors ignored. [Op:__inference_distributed_function_95797]

Errors may have originated from an input operation.
Input Source operations connected to node sequential_12/embedding_12/embedding_lookup:
 sequential_12/embedding_12/embedding_lookup/92192 (defined at /usr/lib/python3.6/contextlib.py:81)

Input Source operations connected to node sequential_12/embedding_12/embedding_lookup:
 sequential_12/embedding_12/embedding_lookup/92192 (defined at /usr/lib/python3.6/contextlib.py:81)

Function call stack:
distributed_function -> distributed_function

Данные 

data['title'] = [['Sentence'],['Sentence2'], ...]

Подготовка данных 

tokenizer = keras.preprocessing.text.Tokenizer(num_words=209, lower=False, char_level=True)
tokenizer.fit_on_texts(df['title'])
df['encoded_with_keras'] = tokenizer.texts_to_sequences(df['title'])

dataset = df['encoded_with_keras']
dataset = tf.keras.preprocessing.sequence.pad_sequences(dataset, padding='post')

dataset = dataset.flatten()

dataset = tf.data.Dataset.from_tensor_slices(dataset)

sequences = dataset.batch(seq_len+1, drop_remainder=True)


def create_seq_targets(seq):
    input_txt = seq[:-1]
    target_txt = seq[1:]
    return input_txt, target_txt

dataset = sequences.map(create_seq_targets)

batch_size = 128

buffer_size = 10000

dataset = dataset.shuffle(buffer_size).batch(batch_size, drop_remainder=True)

Модель: 

vocab_size = len(tokenizer.word_index)
embed_dim = 128
rnn_neurons = 256

epochs = 5

# Directory where the checkpoints will be saved
checkpoint_dir = './training_checkpoints'
# Name of the checkpoint files
checkpoint_prefix = os.path.join(checkpoint_dir, "ckpt_{epoch}")

checkpoint_callback=tf.keras.callbacks.ModelCheckpoint(
    filepath=checkpoint_prefix,
    save_weights_only=True)

def create_model(vocab_size, embed_dim, rnn_neurons, batch_size):
    model = Sequential()
    model.add(Embedding(vocab_size, embed_dim, batch_input_shape=[batch_size, None], mask_zero=True))
    model.add(LSTM(rnn_neurons, return_sequences=True, stateful=True))
    model.add(Dropout(0.2))
    model.add(LSTM(rnn_neurons, return_sequences=True, stateful=True))
    model.add(Dropout(0.2))
    model.compile(optimizer='adam', loss="sparse_categorical_crossentropy")
    return model

model.fit(dataset, epochs=epochs, callbacks=[checkpoint_callback])

Я попытался изменить почти все настройки модели и поиграть с пользовательским токенизацией и подготовкой данных. Но начинается тренировка, и на втором шаге 155 я получаю эту ошибку. Я не уверен, с чего начать, любая помощь приветствуется

...