Tyxe/ezbnn contributions and help

I want to become a Bayesian Neural Network expert! I really would like to contribute to the tyxe library. Can anyone let me know more about it. How I can contribute, learn more about bnn’s and also how to properly use the current tyxe methods. If this is not the right place to ask, my advanced apologies.

Thanks!

Aidan

Hi there,

I appreciate your interest and will surely find things for you to contribute on TyXe.
It is currently still a bit messy due to being in pre-release and I apologize for the many sharp edges, but we plan to have it more stable and polished before ProbProg later this year and you could certainly be our first user and contributor.

Feel free to ping me whenever you have questions about TyXe and please ping me again if I don’t reach out to you in the next few weeks to discuss contributions etc.

Many thanks for your interest.

T.

1 Like

Thank you so much this is sound like a promising lead! I have gotten your resnet example working. But I’m confused about the top5 accuracy. And what metrics I can get out of the model. For example when we preform topk Just so you know the example file for resnet includes ez_bnn in pyro.contrib not tyxe.

https://colab.research.google.com/drive/1brBQKkXgbjRTBX8LB_6zHhFrCsAfuph5?usp=sharing

I’ve commented the confusion.

Thanks!

Aidan

after an hour of careful thinking i have reinvented the top5 acc metric and am currently pursuing wheel improvements. Joking aside, moving onto bert now, more theoretical input would be awesome. Do you ever have a chance to zoom and touch base or something similar. I’m trying to crunch a project into a small amount of time for school. (Creative project) I’d love to assist you deploy this fantastic library and learn as much about bnn in the process. Its a natural idea I’m surprised hasn’t found popularity.

sorry to continually pester you and thank you for your valuable time i just wanted to let you know that the particular issue has been resolved and now I’m faced with a whole host of new considerations – many deep.