关于培训师的混乱。检验（）结果
我想知道以下结果是否不寻常。该值为负，据我所知，低价值表示更好的结果，但只是想确认。
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Testing: 100%
19/19 [00:02<00:00, 9.33it/s]
--------------------------------------------------------------------------------
DATALOADER:0 TEST RESULTS
{'test_loss': -2.5980350971221924}
--------------------------------------------------------------------------------
Out[51]:
[{'test_loss': -2.5980350971221924}]
I wanted to know if the following result is unusual. The value is negative and as far as I know, low values signify better results but just wanted to confirm.
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
Testing: 100%
19/19 [00:02<00:00, 9.33it/s]
--------------------------------------------------------------------------------
DATALOADER:0 TEST RESULTS
{'test_loss': -2.5980350971221924}
--------------------------------------------------------------------------------
Out[51]:
[{'test_loss': -2.5980350971221924}]
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（1）
如果您正在最小化损失，那么是的...较低！
绝对损失值如果您不给出上下文 ie 的功能表达，则没有任何意义。您可以使任何目标函数负面负面保持不变！
If you are minimizing a loss, then yes... lower is better!
Absolute loss values have no meaning if you are not giving context i.e. the functional expression of your objective. You can make any objective function negative while not changing the behaviour of training: you just need to subtract by something big enough and your loss value will end up negative but its derivative will remain unchanged!