Dependency Parser (Pytorch)

最新推荐文章于 2023-03-07 10:58:12 发布

原创

最新推荐文章于 2023-03-07 10:58:12 发布 · 932 阅读

2 ·

CC 4.0 BY-SA版权

因为上一次作业已经写过Logistic Regression了，这次就写parser好了。原理都是差不多的，主要是加了dropout。

class ParserModel(nn.Module):
	def __init__(self, config, word_embeddings=None, pos_embeddings=None,dep_embeddings=None):
		super(ParserModel, self).__init__()
		self.config = config
		n_w = config.word_features_types # 18
		n_p = config.pos_features_types # 18
		n_d = config.dep_features_types # 12
		
		self.word_embeddings = word_embeddings 
		self.pos_embeddings = pos_embeddings # TODO
		self.dep_embeddings = dep_embeddings
		
		self.layer1 = nn.Linear((n_w + n_p + n_d)* self.config.embedding_dim, self.config.l1_hidden_size)	# 2400, 200
		
		self.dropout = nn.Dropout(self.config.keep_prob)
		 
		self.outlayer = nn.Linear(self.config.l1_hidden_size, self.config.num_classes)

		self.init_weights()
		
	def init_weights(self):
		
		initrange = 0.01
		self.layer1.weight.data.uniform_(-initrange, initrange)
		self.layer1.bias.data.zero_()
		self.outlayer.weight.data.uniform_(-initrange, initrange)
		self.outlayer.bias.data.zero_()
		

	def cubic(self,</