深度学习报错：ValueError: Shapes (None,) and (None, 309, 3) are incompatible报错如何解决

Question

我在位bert添加层时报错：
ValueError: Shapes (None,) and (None, 309, 3) are incompatible

这是我的代码：

def load_texts(file_path):
    with open(file_path, 'r', encoding='utf-8') as f:
        texts = [line.strip() for line in f]
    return texts

# 加载数据
AD_texts = load_texts('data/0.txt')
HC_texts = load_texts('data/1.txt')
MCI_texts = load_texts('data/2.txt')

# 分词
AD_texts_tokenized = [' '.join(jieba.cut(text)) for text in AD_texts]
HC_texts_tokenized = [' '.join(jieba.cut(text)) for text in HC_texts]
MCI_texts_tokenized = [' '.join(jieba.cut(text)) for text in MCI_texts]

# 构建数据列表
texts = AD_texts_tokenized + HC_texts_tokenized + MCI_texts_tokenized
labels = [0] * len(AD_texts) + [1] * len(HC_texts) + [2] * len(MCI_texts)

# 对数据进行切分
train_texts, val_texts, train_labels, val_labels = train_test_split(texts, labels, test_size=0.2, random_state=42)



train_encodings = tokenizer(train_texts, truncation=True, padding=True, max_length=512)
val_encodings = tokenizer(val_texts, truncation=True, padding=True, max_length=512)


# 创建tf.data.Dataset
train_dataset = tf.data.Dataset.from_tensor_slices((dict(train_encodings), train_labels))
val_dataset = tf.data.Dataset.from_tensor_slices((dict(val_encodings), val_labels))


train_encodings = tokenizer(train_texts, truncation=True, padding=True, max_length=512)
val_encodings = tokenizer(val_texts, truncation=True, padding=True, max_length=512)

# 创建tf.data.Dataset
train_dataset = tf.data.Dataset.from_tensor_slices((dict(train_encodings), train_labels))
val_dataset = tf.data.Dataset.from_tensor_slices((dict(val_encodings), val_labels))

if 1:
    # 定义BERT模型并抽取特征
    bert_model = TFBertModel.from_pretrained('bert-base-chinese')
    
    input_ids = Input(shape=(309,), dtype=tf.int32, name='input_ids')  # 假设最大长度为512
    attention_mask = Input(shape=(309,), dtype=tf.int32, name='attention_mask')

    sequence_output = bert_model([input_ids, attention_mask])[0]

    # 添加双向LSTM层
    lstm_layer = Bidirectional(LSTM(units=32, return_sequences=True))  # LSTM单元数可自定义
    lstm_out = lstm_layer(sequence_output)

    # 添加全连接层（线性层）
    dense_layer = Dense(units=16, activation='relu')###################
    dense_out = dense_layer(lstm_out)

    # 添加输出层与softmax激活函数
    output_layer = Dense(units=3, activation='softmax')  # 输出类别数为3
    output = output_layer(dense_out)

    # 构建完整模型
    model = Model(inputs=[input_ids, attention_mask], outputs=output)

    # 编译模型
    model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=2e-5), 
                  loss='categorical_crossentropy', 
                  metrics=['accuracy'])

    # 训练模型
    history = model.fit(train_dataset.shuffle(100).batch(1), epochs=2, validation_data=val_dataset.batch(1))

    # 保存模型
    model.save('my_model_add.h5')

我打印的模型结构为：

Model: "model"
__________________________________________________________________________________________________
 Layer (type)                   Output Shape         Param #     Connected to
==================================================================================================
 input_ids (InputLayer)         [(None, 309)]        0           []

 attention_mask (InputLayer)    [(None, 309)]        0           []

 tf_bert_model (TFBertModel)    TFBaseModelOutputWi  102267648   ['input_ids[0][0]',
                                thPoolingAndCrossAt               'attention_mask[0][0]']
                                tentions(last_hidde
                                n_state=(None, 309,
                                 768),
                                 pooler_output=(Non
                                e, 768),
                                 past_key_values=No
                                ne, hidden_states=N
                                one, attentions=Non
                                e, cross_attentions
                                =None)

 bidirectional (Bidirectional)  (None, 309, 64)      205056      ['tf_bert_model[0][0]']

 dense (Dense)                  (None, 309, 16)      1040        ['bidirectional[0][0]']

 dense_1 (Dense)                (None, 309, 3)       51          ['dense[0][0]']

==================================================================================================
Total params: 102,473,795
Trainable params: 102,473,795
Non-trainable params: 0
__________________________________________________________________________________________________

Answer 1

【以下回答由 GPT 生成】

很抱歉，由于你没有给出具体的问题，我无法提供解决方案。请提供具体的问题和上下文，我会尽力帮助你解决问题。

如果你已经解决了该问题, 非常希望你能够分享一下解决方案, 写成博客, 将相关链接放在评论区, 以帮助更多的人 ^-^