WHLYA / text-classification

29 stars 16 forks source link

模型无法收敛 #3

Open xiaomi1213 opened 3 years ago

xiaomi1213 commented 3 years ago

你好,我用您的textcnn的模型做回归问题,但是无法收敛,同样的网络结构在Keras却可以收敛

WHLYA commented 3 years ago

和框架没关系吧。你改了代码哪里吗?

------------------ 原始邮件 ------------------ 发件人: "WHLYA/text-classification" <notifications@github.com>; 发送时间: 2020年12月14日(星期一) 中午11:30 收件人: "WHLYA/text-classification"<text-classification@noreply.github.com>; 抄送: "Subscribed"<subscribed@noreply.github.com>; 主题: [WHLYA/text-classification] 模型无法收敛 (#3)

你好,我用您的textcnn的模型做回归问题,但是无法收敛,同样的网络结构在Keras却可以收敛

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.

xiaomi1213 commented 3 years ago

基本没怎么改,就加了BatchNorm1d,还有就是改了一维卷积的kernel_sizes = [3,4,5],num_channels = [256,256,256],其他的一样

WHLYA commented 3 years ago

原来的是可以运行的吗

------------------ 原始邮件 ------------------ 发件人: "xiaomi1213"<notifications@github.com>; 发送时间: 2020年12月14日(星期一) 中午12:24 收件人: "WHLYA/text-classification"<text-classification@noreply.github.com>; 抄送: "青青草原"<2647480632@qq.com>; "Comment"<comment@noreply.github.com>; 主题: Re: [WHLYA/text-classification] 模型无法收敛 (#3)

基本没怎么改,就加了BatchNorm1d,还有就是改了一维卷积的kernel_sizes = [3,4,5],num_channels = [256,256,256],其他的一样

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

xiaomi1213 commented 3 years ago

可以运行是可以运行,就是损失函数不收敛,我是用来做回归问题,用的损失函数是mse_loss