/TorchAssembly

Simple utilities to allow for automated or queued hyper-parameter optimisation in PyTorch

Primary LanguagePython

Desired Features:

  • Support Default Values TICK
  • JSON Format TICK
    • Save/Load as JSON (can edit directly from file) TICK
    • Allow for conversion to nested Kwargs TICK
  • Support Synonyms
  • Shared Characteristics (only define same hyper-parameter once) TICK
  • Define hyper-parameters dynamically TICK
    • Add new hyper-parameter to hyper-parameters instance, optionally parent blueprint TICK
  • Parameterise the setup process (smaller function signatures)
  • Add thorough documentation
  • Use same Saving Convention for both Blueprint and Hyperparameters TICK
  • Add Logging TICK
  • Add Argparse Control TICK
  • Add UnitTesting TICK
  • Add GitHub build status