python - ValueError：维度必须为 4，但对于输入形状为 [?,10,16,16], [3] 的“capsule_2/transpose_4”（操作：“Transpose”）为 3

Question

我正在尝试使用胶囊网络进行二进制文本分类。我使用预训练的 fasttext 词嵌入来矢量化我的数据集

，这是我的模型定义

 gru_len = 128
    Routings = 5
    Num_capsule = 10
    Dim_capsule = 16
    dropout_p = 0.
    rate_drop_dense = 0.3

def get_model(embedding_matrix, sequence_length, dropout_rate, recurrent_units, dense_size):
    input1 = Input(shape=(sequence_length,))
    embed_layer = Embedding(embedding_matrix.shape[0], embedding_matrix.shape[1],
                                weights=[embedding_matrix], trainable=False)(input1)
    embed_layer = SpatialDropout1D(rate_drop_dense)(embed_layer)

    x = Bidirectional(
        GRU(gru_len, activation='relu', dropout=dropout_p, recurrent_dropout=dropout_p, return_sequences=True))(embed_layer)
    capsule = Capsule(num_capsule=Num_capsule, dim_capsule=Dim_capsule,routings=Routings,share_weights=True)(x)
    capsule = Flatten()(capsule)
    capsule = Dropout(dropout_p)(capsule)
    output = Dense(1, activation='sigmoid')(capsule)
    model = Model(inputs=input1, outputs=output)
    model.compile(
        loss='binary_crossentropy',
        optimizer='adam',
        metrics=['accuracy'])
    model.summary()
    return model

但是当模型开始训练时，我得到了一个错误。错误日志如下：

ValueError: Dimension must be 4 but is 3 for 'capsule_3/transpose_4' (op: 'Transpose') with input shapes: [?,10,16,16], [3].

标签： pythonmachine-learningtext-classification