running notebook with custom loss class for skorch.

openkinome / kinoml

Structure-informed machine learning for kinase modeling

https://openkinome.org/kinoml/

MIT License

51 stars 21 forks source link

running notebook with custom loss class for skorch. #27

Open t-kimber opened 4 years ago

t-kimber commented 4 years ago

Description

This PR focuses on how to use skorch (Pytorch + sklearn) package for training the machine learning models. Advantages: Cool features pre-implemented, e.g.

Built in cross-validation
early stopping
callbacks
etc.

Todos

[ ] Include skorch in API
[x] running example
- [x] custom loss class for pIC50 to enable observation model

Questions

[ ] Question1

Status

[ ] Ready to go

jaimergp commented 4 years ago

This is so cool!