nf-core / deepmodeloptim

Stochastic Testing and Input Manipulation for Unbiased Learning Systems
https://nf-co.re/deepmodeloptim
MIT License
23 stars 9 forks source link

Train module #72

Closed alessiovignoli closed 7 months ago

alessiovignoli commented 7 months ago

created the subworflow the module necessary for the training of torch models. Changed the way experiments are loaded and passed to ray.tune. Changed the way the model is loaded and passed to ray.tune that will then initialize it. Fixed ray.tune not able to open relative paths inside workdir of nextflow. Adapted model class in test to fit new import strategy from launc_training.py.

TODO :

  1. better creation of subdirs specific to each process of torch:train, to stroe bestmodel and bestconfig
  2. find a way to retrieve bestmodel from ray.tune