Question

我正在使用带有TensorFlow Hub的Keras的Lambda层从预建的嵌入器中下载单词嵌入。

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)

我希望Keras能够建立一个包含输入中每个单词128个单词嵌入列表的模型。实际上，Lambda层在“嵌入”功能上运行时会导致以下错误。

“ ValueError：Tensor转换请求具有dtype float32的Tensor的dtype字符串：'Tensor（” sequential_5_input：0“，shape =（?, 2），dtype = float32）'”

根据GitHub问题（https://github.com/keras-team/keras/issues/10021上的用户nuric），此问题是由Keras尝试推断输出形状引起的。如您所见，我尝试通过指定所需的输出形状来解决此问题。

这是神经网络的输入和所需的输出：

输入

[['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.']

所需的输出：

[[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.]]

Answer 1

我刚刚尝试过，当我删除“ input_shape = [None]”时，它对我有用。所以这段代码应该可以工作：

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()

编辑

此keras模型应等于SequentialModel（显式输入层除外）：

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(...)

运行此命令对我有用...

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

...在lambda函数中对此进行更改后：

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

使用dtype float32（lambda输入）的Tensor的Tensor转换请求dtype字符串

1 个答案: