Skip to content

为什么要在预测阶段随机初始化hidden层? #1

Description

@pustar

您好!感谢分享。有个问题,每次测试结果在一定范围随机变化,这不合理。即使model.eval()关掉dropout,也依然随机。原因是:

    def init_hidden(self):
        return (torch.randn(2, self.batch_size, self.hidden_dim // 2),
                torch.randn(2, self.batch_size, self.hidden_dim // 2))
    def _get_lstm_features(self, sentences):
        """BiLSTM的输出(还没经过CRF层)"""
        self.hidden = self.init_hidden()

为什么要在预测阶段采用随机初始化?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions