Fastai

New Features

- Add support for pytorch 1.8 ([3349](https://github.com/fastai/fastai/issues/3349))
- Add support for spacy3 ([3348](https://github.com/fastai/fastai/issues/3348))
- Add support for Windows. Big thanks to Microsoft for many contributions to get this working
- Timedistributed layer and Image Sequence Tutorial ([3124](https://github.com/fastai/fastai/pull/3124)), thanks to [tcapelle](https://github.com/tcapelle)
- Add interactive run logging to AzureMLCallback ([3341](https://github.com/fastai/fastai/pull/3341)), thanks to [yijinlee](https://github.com/yijinlee)
- Batch support for more than one image ([3339](https://github.com/fastai/fastai/issues/3339))
- Have interp use ds_idx, add tests ([3332](https://github.com/fastai/fastai/pull/3332)), thanks to [muellerzr](https://github.com/muellerzr)
- Automatically have fastai determine the right device, even with torch DataLoaders ([3330](https://github.com/fastai/fastai/pull/3330)), thanks to [muellerzr](https://github.com/muellerzr)
- Add `at_end` feature to `SaveModelCallback` ([3296](https://github.com/fastai/fastai/pull/3296)), thanks to [tmabraham](https://github.com/tmabraham)
- Improve inplace params in Tabular's new and allow for new and test_dl to be in place ([3292](https://github.com/fastai/fastai/pull/3292)), thanks to [muellerzr](https://github.com/muellerzr)
- Update VSCode & Codespaces dev container ([3280](https://github.com/fastai/fastai/pull/3280)), thanks to [bamurtaugh](https://github.com/bamurtaugh)
- Add max_scale param to RandomResizedCrop(GPU) ([3252](https://github.com/fastai/fastai/pull/3252)), thanks to [kai-tub](https://github.com/kai-tub)
- Increase testing granularity for speedup ([3242](https://github.com/fastai/fastai/pull/3242)), thanks to [ddobrinskiy](https://github.com/ddobrinskiy)

Bugs Squashed

- Make TTA turn shuffle and drop_last off when using ds_idx ([3347](https://github.com/fastai/fastai/pull/3347)), thanks to [muellerzr](https://github.com/muellerzr)
- Add order to TrackerCallback derived classes ([3346](https://github.com/fastai/fastai/pull/3346)), thanks to [muellerzr](https://github.com/muellerzr)
- Prevent schedule from crashing close to the end of training ([3335](https://github.com/fastai/fastai/pull/3335)), thanks to [Lewington-pitsos](https://github.com/Lewington-pitsos)
- Fix ability to use raw pytorch DataLoaders ([3328](https://github.com/fastai/fastai/pull/3328)), thanks to [hamelsmu](https://github.com/hamelsmu)
- Fix PixelShuffle_icnr weight ([3322](https://github.com/fastai/fastai/pull/3322)), thanks to [pratX](https://github.com/pratX)
- Creation of new DataLoader in Learner.get_preds has wrong keyword ([3316](https://github.com/fastai/fastai/pull/3316)), thanks to [tcapelle](https://github.com/tcapelle)
- Correct layers order in tabular learner ([3314](https://github.com/fastai/fastai/pull/3314)), thanks to [gradientsky](https://github.com/gradientsky)
- Fix vmin parameter default ([3305](https://github.com/fastai/fastai/pull/3305)), thanks to [tcapelle](https://github.com/tcapelle)
- Ensure call to `one_batch` places data on the right device ([3298](https://github.com/fastai/fastai/pull/3298)), thanks to [tcapelle](https://github.com/tcapelle)
- Fix Cutmix Augmentation ([3259](https://github.com/fastai/fastai/pull/3259)), thanks to [MrRobot2211](https://github.com/MrRobot2211)
- Fix custom tokenizers for DataLoaders ([3256](https://github.com/fastai/fastai/pull/3256)), thanks to [iskode](https://github.com/iskode)
- fix error setting 'tok_tfm' parameter in TextDataloaders.from_folder
- Fix lighting augmentation ([3255](https://github.com/fastai/fastai/pull/3255)), thanks to [kai-tub](https://github.com/kai-tub)
- Fix CUDA variable serialization ([3253](https://github.com/fastai/fastai/pull/3253)), thanks to [mszhanyi](https://github.com/mszhanyi)
- change batch tfms to have the correct dimensionality ([3251](https://github.com/fastai/fastai/pull/3251)), thanks to [trdvangraft](https://github.com/trdvangraft)
- Ensure add_datepart adds elapsed as numeric column ([3230](https://github.com/fastai/fastai/pull/3230)), thanks to [aberres](https://github.com/aberres)

Breaking Changes

- fix optimwrapper to work with `param_groups` ([3241](https://github.com/fastai/fastai/pull/3241)), thanks to [tmabraham](https://github.com/tmabraham)
- OptimWrapper now has a different constructor signature, which makes it easier to wrap PyTorch optimizers

New Features

- Support discriminative learning with OptimWrapper ([2829](https://github.com/fastai/fastai/issues/2829))

Bugs Squashed

- Updated to support adding transforms to multiple dataloaders ([3268](https://github.com/fastai/fastai/pull/3268)), thanks to [marii-moe](https://github.com/marii-moe)
- This fixes an issue in 2.2.7 which resulted in incorrect validation metrics when using Normalization

Bugs Squashed

- Regression fix: Ensure `add_datepart` adds elapsed as numeric column ([3230](https://github.com/fastai/fastai/pull/3230)), thanks to [aberres](https://github.com/aberres)

Bugs Squashed

New Features

- Enhancement: Let TextDataLoaders take in a custom `tok_text_col` ([3208](https://github.com/fastai/fastai/pull/3208)), thanks to [muellerzr](https://github.com/muellerzr)
- Changed dataloaders arguments to have consistent overrides ([3178](https://github.com/fastai/fastai/pull/3178)), thanks to [marii-moe](https://github.com/marii-moe)
- Better support for iterable datasets ([3173](https://github.com/fastai/fastai/pull/3173)), thanks to [jcaw](https://github.com/jcaw)

Bugs Squashed

- BrokenProcessPool in `download_images()` on Windows ([3196](https://github.com/fastai/fastai/issues/3196))
- error on predict() or using interp with resnet and MixUp ([3180](https://github.com/fastai/fastai/issues/3180))
- Fix 'cat' attribute with pandas dataframe: `AttributeError: Can only use .cat accessor with a 'category' dtype` ([3165](https://github.com/fastai/fastai/pull/3165)), thanks to [dreamflasher](https://github.com/dreamflasher)
- `cont_cat_split` does not support pandas types ([3156](https://github.com/fastai/fastai/issues/3156))
- `DataBlock.dataloaders` does not support the advertised "shuffle" argument ([3133](https://github.com/fastai/fastai/issues/3133))

New Features

- Calculate correct `nf` in `create_head` based on `concat_pool` ([3115](https://github.com/fastai/fastai/pull/3115)), thanks to [muellerzr](https://github.com/muellerzr)

Bugs Squashed

- wandb integration failing with latest wandb library ([3066](https://github.com/fastai/fastai/issues/3066))
- `Learner.load` and `LRFinder` not functioning properly for the optimizer states ([2892](https://github.com/fastai/fastai/issues/2892))