search

SpyrosMouselinos / BRITSTF2

A Tensorflow 2 Implementation of the NEURIPS 2018 BRITS Paper
Apache License 2.0
2 stars 0 forks source link

Forward and Backward Imputation Mean #1

Open hasnainroopawalla opened 2 years ago

hasnainroopawalla commented 2 years ago

Is this line missing here in your BRITS implementation here?

The paper (page 5) mentions: The final estimation in the t-th step is the mean of xˆt and xˆt'.

Should there be a mean of the forward and backward imputations here?

SpyrosMouselinos commented 2 years ago

You have written the same term twice xˆt, so I am not sure what you are referring to. Nevertheless, the project here is kind of old and not maintained actively by me anymore, if you have any suggestions or updates you can to a pull request to improve it.

hasnainroopawalla commented 2 years ago

That was my mistake, I have now fixed my comment. I'll put up a PR soon!

sayantanghosh commented 1 year ago

That was my mistake, I have now fixed my comment. I'll put up a PR soon!

Hi @hasnainroopawalla: Did you update the code, or have you tested your suggestion?

hasnainroopawalla commented 1 year ago

Hi @sayantanghos, I abandoned using the BRITS model and stuck with the RITS model. However, from what I understand, the mean of the backward imputations should be computed too.

sayantanghosh commented 1 year ago

@hasnainroopawalla: that's what I figured from another careful reading of the BRITS paper. Thank you for your reply.