Skip to content

Releases: wolfhoundgelert/xman

Fix

11 Jun 09:55
Compare
Choose a tag to compare
Fix Pre-release
Pre-release

Fix for AttributeError: module 'xman.platform' has no attribute 'check_jupyter_notebook_kernel_interrupted'

Release 0.2.0

06 Jun 20:19
1809ffd
Compare
Choose a tag to compare
Release 0.2.0 Pre-release
Pre-release

New functional, improvements, bugfixing:

  • "Exp 1 [DONE]" -> "Exp 2.1 [DONE]" (add group number)

  • For exp.success() and exp.fail() check there's no other success or fail status - need that guard

  • Can't unpickle exp after re-oping in other notebook (DL2, hw_6-1, Group 1, Exp 1 and 2)

    • Try to use dill instead of cloudpickle - NO NEED, the issue was about datasets load its data from disk.
      In such cases it's better to add to the train run-func all needed code:
      !pip install datasets
      import datasets
      dataset = datasets.load_dataset('ag_news')
      
      import nltk
      nltk.download('punkt')
      
  • Opened under another account project doesn't recognize exp running in the first group as IN_PROGRESS and ACTIVE. Shows it as TO_DO. (exp 3.1 in DL2/hw_6-1)

  • Create exp with saving on storage, run exp and stop it manually in the cell. Start once again:

    IllegalOperationXManError: Can't recreate pipeline for exp `Exp 2 [IN_PROGRESS: IDLE] Baseline `mean` - {'aggregation_type': 'mean'}` - there's no `.run` data file! Use `save=True` for `make_pipeline()` method if you need to preserve `run_func` and `params` for other session.
    
    • Maybe it would be better if I didn't delete .run after an error, because I can tune some in imports and run. But need to delete error and error_stack on rerun, and review start().
  • Make 2 exp-s, xman.start(). The second exp started, but the first one was expected.

  • Possibility to mark some exp with some marker in info(). I want to mark best or interesting exps. And clean that marker if it's needed.

  • What For and Why section before Installation:

    • Case 1: GPU, checkpoints, spent hours for nothing when GPU was dropped by Google
      +Case 2: Chaotic research without understanding
  • Add a note about issues with pickling to the Running in parallel... section

  • Add to the readme a list of main functionality (commonly used methods of API)

Pilot

01 Jun 09:32
Compare
Choose a tag to compare
Pilot Pre-release
Pre-release

Pilot release