Improve save_path in AutoMM #2635
Conversation
|
Job PR-2635-af78a0e is done. |
|
Job PR-2635-6c05703 is done. |
|
Job PR-2635-6a33bad is done. |
| save_path = process_save_path(path=old_save_path, raise_if_exist=False) | ||
| else: | ||
| save_path = os.path.abspath(os.path.expanduser(old_save_path)) | ||
| save_path = process_save_path(path=old_save_path, raise_if_exist=(raise_if_exist and rank == 0)) |
There was a problem hiding this comment.
We may need a multi-GPU machine in the CI to test for multi-GPU training next time.
There was a problem hiding this comment.
Agree. Maybe select partial tests to avoid slow down CI.
|
Job PR-2635-51206e7 is done. |
Issue #, if available:
.predict()would create a new save folder and change_save_path.init()complicates the logic of save_path.Description of changes:
init(). Create saving folder only when it's used._model_loadedflag. For saving the prediction results of object detection, we can directly save to the path_save_path, whether the model is loaded or fitted.By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.