mindtext.modules.encoder.conv

class mindtext.modules.encoder.conv.ConvEncoder (init_embed: StaticEmbedding, num_filters: int = 256, kernel_size: int = 3, num_layers: int = 7, embed_dropout: float = 0.1)

卷积编码器

Example

>>> vocab = Vocabulary()
>>> vocab.update(["i", "am", "fine"])
>>> embed = StaticEmbedding(vocab, model_dir_or_name=None, embedding_dim=100)
>>> conv_encoder = ConvEncoder(embed)
>>> words = mindspore.Tensor(np.random.randint(0, 3, (1, 256)))
>>> x = conv_encoder(words)

init (init_embed: StaticEmbedding, num_filters: int = 256, kernel_size: int = 3, num_layers: int = 7, embed_dropout: float = 0.1)

参数

• init_embed (StaticEmbedding): StaticEmbedding。

• num_filters (int): 过滤器的数量，默认为256。

• kernel_size (int): 卷积核的大小，默认为3。

• num_layers (int): CNN的复合数量，默认为7。

• embed_dropout (float): dropout层的概率，默认为0.1。

construct (word)

参数

• word (Tensor): 输入的向量。

返回

• x (Tensor): 卷积编码层返回的向量。