Tensor-преобразование запросило строку dtype для Tensor с dtype float32 (лямбда-вход) - PullRequest
Новая
       12

Tensor-преобразование запросило строку dtype для Tensor с dtype float32 (лямбда-вход)

0 голосов
/ 05 апреля 2019

Я использую лямбда-слой Keras с TensorFlow Hub для загрузки вложений слов из предварительно созданного встроенного устройства. 

import tensorflow_hub as hub
from tensorflow.dtypes import as_string
def embedding(x):
    print(x.shape)
    module = hub.Module("https://tfhub.dev/google/nnlm-en-dim128/1")
    return module(x)
answers_network_rnn = Sequential()
print(trainingData["question"].shape)
answers_network_rnn.add(Lambda(embedding,output_shape=(128,)))
answers_network_rnn.add(Dense(16))
answers_network_rnn.add(Dense(Y_2_train_num.shape[1]))
answers_network_rnn.summary()
answers_network_rnn.compile("adam","categorical_crossentropy",metrics=['accuracy',f1]) answers_network_rnn_checkpoint = ModelCheckpoint('answers_network-rnn-best.h5', verbose=1, monitor='val_f1',save_best_only=True, mode='auto') answers_network_rnn.fit(x=X_2_train_text.values,y=Y_2_train_num)

Я ожидаю, что Keras создаст модель со списком из 128 вложений слов для каждогоСлово в моих входах.На самом деле, лямбда-слой вызывает следующую ошибку при запуске функции «встраивание».

«ValueError: Запрошено преобразование Tensor строки dtype для Tensor с dtype float32: 'Tensor (" sequential_5_input: 0 ", shape= (?, 2), dtype = float32) '"

В соответствии с пользовательским nuric по проблеме GitHub (https://github.com/keras-team/keras/issues/10021) эта проблема вызвана тем, что Keras пытается вывести форму вывода. Как вы можетевидите, я попытался решить эту проблему, указав желаемую форму вывода.

Вот вход и желаемый вывод для нейронной сети:

Вход 

[['to whom did the virgin mary allegedly appear in 1858 in lourdes france?'
  'architecturally, the school has a catholic character. atop the main building\'s gold dome is a golden statue of the virgin mary. immediately in front of the main building and facing it, is a copper statue of christ with arms upraised with the legend "venite ad me omnes". next to the main building is the basilica of the sacred heart. immediately behind the basilica is the grotto, a marian place of prayer and reflection. it is a replica of the grotto at lourdes, france where the virgin mary reputedly appeared to saint bernadette soubirous in 1858. at the end of the main drive (and in a direct line that connects through 3 statues and the gold dome), is a simple, modern stone statue of mary.']
 ['what is in front of the notre dame main building?'
  'architecturally, the school has a catholic character. atop the main building\'s gold dome is a golden statue of the virgin mary. immediately in front of the main building and facing it, is a copper statue of christ with arms upraised with the legend "venite ad me omnes". next to the main building is the basilica of the sacred heart. immediately behind the basilica is the grotto, a marian place of prayer and reflection. it is a replica of the grotto at lourdes, france where the virgin mary reputedly appeared to saint bernadette soubirous in 1858. at the end of the main drive (and in a direct line that connects through 3 statues and the gold dome), is a simple, modern stone statue of mary.']
 ['the basilica of the sacred heart at notre dame is beside to which structure?'
  'architecturally, the school has a catholic character. atop the main building\'s gold dome is a golden statue of the virgin mary. immediately in front of the main building and facing it, is a copper statue of christ with arms upraised with the legend "venite ad me omnes". next to the main building is the basilica of the sacred heart. immediately behind the basilica is the grotto, a marian place of prayer and reflection. it is a replica of the grotto at lourdes, france where the virgin mary reputedly appeared to saint bernadette soubirous in 1858. at the end of the main drive (and in a direct line that connects through 3 statues and the gold dome), is a simple, modern stone statue of mary.']

Желаемый выход: 

[[0. 0. 0. ... 0. 0. 0.]
 [0. 0. 0. ... 0. 0. 0.]
 [0. 0. 0. ... 0. 0. 0.]
 ...
 [0. 0. 0. ... 0. 0. 0.]
 [0. 0. 0. ... 0. 0. 0.]
 [1. 0. 0. ... 0. 0. 0.]]

1 Ответ

1 голос
/ 05 апреля 2019

Я только что попробовал это, и оно работает для меня, когда я удаляю "input_shape = [None],". Так что этот код должен работать: 

import tensorflow_hub as hub
from tensorflow.dtypes import as_string
def embedding(x):
    print(x.shape)
    module = hub.Module("https://tfhub.dev/google/nnlm-en-dim128/1")
    return module(x)
answers_network_rnn = Sequential()
print(trainingData["question"].shape)
from keras.layers import InputLayer
answers_network_rnn.add(Lambda(embedding,output_shape=(128,)))
answers_network_rnn.add(Dense(16))
answers_network_rnn.add(Dense(Y_2_train_num.shape[1]))
answers_network_rnn.summary()

EDIT

Эта модель keras должна быть равна SequentialModel (кроме явного входного слоя): 

input_text = tf.keras.layers.Input(shape=(1,), dtype=tf.string)
embedding_layer = tf.keras.layers.Lambda(embedding,output_shape=(128,))(input_text)
dense = tf.keras.layers.Dense(16)(embedding_layer)
outputs = tf.keras.layers.Dense(Y_2_train_num.shape[1])(dense)

answers_network_rnn = tf.keras.Model(inputs=[input_text], outputs=outputs)
answers_network_rnn.compile(...)

Запуск этого работает для меня ... 

with tf.Session() as session:
  session.run([tf.global_variables_initializer(), tf.tables_initializer()])
  answers_network_rnn.fit(...)

... после изменения в лямбда-функции: 

#return module(x)
return module(tf.squeeze(tf.cast(x, tf.string)),signature="default", as_dict=True)["default"]
...