- Change the frequency of Tensorboard logging to every
num_log_iter
iterations. - Aim tracking now resumes tracking the same run when resuming training.
- Simplified and improved the flexibility of the dependencies.
- Fixes hard-coded repository path for Aim experiment tracking.
- Added DeFixmatch algorithm.
- Added imbalanced Semi-Supervised setting and 8 imbalanced algorithms.
- Added FreeMatch and SoftMatch algorithms.
- Added registry mechanism for calling algorithms.
- Added wandb and aimstack support.
- Reformated 'train_step' structure.
- Updated results (and logs) [Google Drive] [One Drive] with online wandb logs [Wandb].
- Stable release semilearn==0.3.0
- Updated Colab tutorial
- Updated dataset download and pre-process instructions
- Reformated 'train_step' structure with component hooks.
- Added training hooks.