antoinecarme/pyaf

Use PyTorch as the reference deep learning framework/architecture for future projects

antoinecarme opened this issue · 2 comments

antoinecarme commented

PyAF will use PyTorch as its deep learning architecture for future projects. A few reasons for this :

  1. Pytorch is fully open source. Green (#176 )
  2. PyTorch internal/technical choices are very sane. It works even in very hash environments : SPARC64 architecture.
  3. SPARC64 architecture : abandoned years ago, no commercial support, very strong technically ( manycore, > 128 threads), with approximate OS (Debian rocks here ;), was able to build a set of packages for PyTorch from scratch : https://github.com/antoinecarme/sparc-t3-data/tree/master/debian-sparc64/packages
  4. PyAF runs OK with PyTorch on SPARC64 and uses all the 128 threads for some complex hierarchical forecasting models.
antoinecarme commented

Get rid of Keras and tensorflow. And now that we have pytorch working (#199 ), no need for scikeras in a pyaf context neither.

Less is better. Minimal is beautiful.

DeprecationWarning: KerasRegressor is deprecated, use Sci-Keras (https://github.com/adriangb/scikeras) instead. See https://www.adriangb.com/scikeras/stable/migration.html for help migrating.
antoinecarme commented

Closing.