How to decide the value of amplification coefficient

[mzy97]

In paper and code implementation, you set the sigmoid function in Logit model need to amplified by 12, how to decide this value? If I change the task to image segmentation, will this value be changed, how to set it?

thanks

[sebgao]

Any large (but not too large) should be OK. But honestly, I tend to remove SoftGate (sigmoid + amplified coefficient) if the model does not meet NaN errors.

…

________________________________ 发件人: mzy97 <notifications@github.com> 发送时间: Thursday, September 26, 2019 10:09:13 AM 收件人: sebgao/LIP <LIP@noreply.github.com> 抄送: Subscribed <subscribed@noreply.github.com> 主题: [sebgao/LIP] How to decide the value of amplification coefficient (#5) In paper and code implementation, you set the sigmoid function in Logit model need to amplified by 12, how to decide this value? If I change the task to image segmentation, will this value be changed, how to set it? thanks — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<#5?email_source=notifications&email_token=AGH4UHKURQJ7HQ3FCYO5K5LQLQKUTA5CNFSM4I2UALH2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HNYGXPQ>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AGH4UHJBBXDF3EDZRZ3KPRLQLQKUTANCNFSM4I2UALHQ>.

[mzy97]

got it, thank you