Question

我尝试使用keras来建立注意力集中的文档分类模型（论文分类注意力网络的一部分）。以下代码是测试代码。我创建了一个框，并引用了https://github.com/person-lee/LSTM_ATTENTION_CLASSIFY/blob/master/utils.py和https://github.com/richliao/textClassifier/blob/master/textClassifierHATT.py的自定义注意层。但是我收到一个错误（请参阅下面的详细信息）。

代码为：

from keras.models import Model
from keras.layers import Input
from keras.layers.embeddings import Embedding
from keras.layers.recurrent import GRU
from keras.layers.wrappers import Bidirectional, TimeDistributed
from keras.layers.core import Dropout, Dense, Lambda, Masking
from keras.layers import merge
from keras.engine.topology import Layer

from keras import backend as K
from keras import initializers
import keras
class AttentionLayer(Layer):
    '''
    Attention layer. 
    '''
    def __init__(self, init='glorot_uniform', **kwargs):
        super(AttentionLayer, self).__init__(**kwargs)
        self.supports_masking = True
        self.init = initializers.get(init)

    def build(self, input_shape):
        input_dim = input_shape[-1]
        self.Uw = self.init((input_dim, ))
        self.trainable_weights = [self.Uw]
        super(AttentionLayer, self).build(input_shape)  

    def compute_mask(self, input, mask):
        return mask

    def call(self, x, mask=None):
        eij = K.tanh(K.squeeze(K.dot(x, K.expand_dims(self.Uw)), axis=-1))
        ai = K.exp(eij)
        weights = ai/K.expand_dims(K.sum(ai, axis=1),1)

        weighted_input = x*K.expand_dims(weights,2)
        return K.sum(weighted_input, axis=1)

    def get_output_shape_for(self, input_shape):
        newShape = list(input_shape)
        newShape[-1] = 1
        return tuple(newShape)

sentence_input = Input(shape=(None,5))
# embedded_sequences = embedding_layer(sentence_input)
l_lstm = Bidirectional(GRU(10, return_sequences=True),merge_mode='concat')(sentence_input)
# l_dense = TimeDistributed(Dense(200))(l_lstm)
l_att = AttentionLayer()(l_lstm)
cls = Dense(10, activation='softmax')(l_att)
sentEncoder = Model(sentence_input, cls)

sentEncoder.compile(loss='categorical_crossentropy',
              optimizer='rmsprop',
              metrics=['acc'])
import numpy as np
x_train = np.array([[1,2,3,4,5],
                    [1,2,3,4,5],
                    [1,2,3,4,5],
                    [1,2,3,4,5],
                    [1,2,3,4,5],
                    [1,2,3,4,5],
                    [1,2,3,4,5],
                    [1,2,3,4,5],
                    [1,2,3,4,5],
                    [1,2,3,4,5]])
y_train = np.array([1,2,3,4,5,6,7,8,9,0])
y_train = keras.utils.to_categorical(y_train, 10)
x_train = np.expand_dims(x_train,0)
y_train = np.expand_dims(y_train,0)

sentEncoder.fit(x=x_train,y=y_train,validation_split=0.1)

并出现以下错误：

AttributeError                            Traceback (most recent call last)
<ipython-input-13-3f6bb30d8618> in <module>()
----> 1 sentEncoder.fit(x=x_train,y=y_train,validation_split=0.1)

~/.conda/envs/21/lib/python3.6/site-packages/keras/engine/training.py in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, **kwargs)
   1011         else:
   1012             ins = x + y + sample_weights
-> 1013         self._make_train_function()
   1014         f = self.train_function
   1015 

~/.conda/envs/21/lib/python3.6/site-packages/keras/engine/training.py in _make_train_function(self)
    495                     training_updates = self.optimizer.get_updates(
    496                         params=self._collected_trainable_weights,
--> 497                         loss=self.total_loss)
    498                 updates = (self.updates +
    499                            training_updates +

~/.conda/envs/21/lib/python3.6/site-packages/keras/legacy/interfaces.py in wrapper(*args, **kwargs)
     89                 warnings.warn('Update your `' + object_name +
     90                               '` call to the Keras 2 API: ' + signature, stacklevel=2)
---> 91             return func(*args, **kwargs)
     92         wrapper._original_function = func
     93         return wrapper

~/.conda/envs/21/lib/python3.6/site-packages/keras/optimizers.py in get_updates(self, loss, params)
    262                 new_p = p.constraint(new_p)
    263 
--> 264             self.updates.append(K.update(p, new_p))
    265         return self.updates
    266 

~/.conda/envs/21/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py in update(x, new_x)
    968         The variable `x` updated.
    969     """
--> 970     return tf.assign(x, new_x)
    971 
    972 

~/.conda/envs/21/lib/python3.6/site-packages/tensorflow/python/ops/state_ops.py in assign(ref, value, validate_shape, use_locking, name)
    282         ref, value, use_locking=use_locking, name=name,
    283         validate_shape=validate_shape)
--> 284   return ref.assign(value, name=name)
    285 
    286 

AttributeError: 'Tensor' object has no attribute 'assign'

我不知道怎么了。我用谷歌搜索并询问擅长此事的人，但没有弄清楚。是因为bidirectional吗？有人知道出了什么问题吗？请告诉我！预先感谢！

Answer 1

我想这是数据集和标签形状的问题。

Answer 2

我遇到了同样的问题，我解决了。原因是K.update（p，new_p），'p'类型不应该是张量类型，当您使用K.update（p，new_p），'p'类型应该是tf.Variable和'new_p'类型应该是张量类型，我希望它可以解决您的问题。

AttributeError：注意模型中的“ Tensor”对象没有属性“ assign”

2 个答案: