v0.2.0

What's Changed

Other Changes

• Apparently "!*" has some meaning in bash by @tmke8 in #212
• Modernize some enums, fix some other things by @tmke8 in #213
• Restructure the CI by @tmke8 in #214
• Use python 3.8 in CI by @tmke8 in #215
• Port iWildCam and fMoW datasets from WILDS by @MylesBartlett in #217
• Fix errors introduced by conversion to StrEnum. by @MylesBartlett in #219
• Format with experimental settings in black by @tmke8 in #220
• Try to make audio data processing work even if Albumentations is not installed by @tmke8 in #221
• Remove final annotation from CdtDataset.getitem. by @MylesBartlett in #222
• Remove extras specifier from ranzen as ranzen no longer has extras. by @MylesBartlett in #223
• Use type stubs for cv2 by @tmke8 in #224
• Delete relay.py by @tmke8 in #228
• Add option to use 'fixed' version of the Waterbirds dataset introduced in MaskTune. by @MylesBartlett in #230
• Set up auto-approve and auto-merge for dependabot by @tmke8 in #235
• Make directory for Waterbirds dataset + (fixed) csv file and fix filepath. by @MylesBartlett in #236
• Try to fix dependabot-automerge by @tmke8 in #241
• Make fscore robust to disjointness in the predicted and target labels. by @MylesBartlett in #247
• Allow split argument of CMNIST to encode a custom split. by @MylesBartlett in #127
• Add option to include unsupported predicted labels in (as is done in sklearn). by @MylesBartlett in #248
• Replace torch._six.string_classes with str by @tmke8 in #273
• Make compatible with ranzen 2.1 by @tmke8 in #274
• Update torch and use **kwargs for the DataLoader subclass by @tmke8 in #275
• Raise highest supported version to Python 3.11 by @tmke8 in #278
• Update pytorch lightning by @tmke8 in #282

Full Changelog: 0.1.7...v0.2.0

0.1.7

What's Changed

Bug fixes 🐞

• Clone 'seed' ids for index_set in groupwise_metric to prevent in-place modification. by @MylesBartlett in #173

Other Changes

• Raise min versions of neoconfigen (for Union/Literal support) and torch (for new scatter_reduce interface). by @MylesBartlett in #175
• Create dependabot.yml by @olliethomas in #176
• Add option to return raw indices to stratified_split. by @MylesBartlett in #179
• Set exclude_self in reduce_scatter to False to correct 1-off error. by @MylesBartlett in #187
• Add fscore and subclass/group-conditional variants to metrics. by @MylesBartlett in #192
• Update EthicML by @thomkeh in #193
• Add NIH Chest X-rays dataset. by @MylesBartlett in #209

Full Changelog: v0.1.5...0.1.7

v0.1.6

What's Changed

Other Changes

• Clone 'seed' ids for index_set in groupwise_metric to prevent in-place modification. by @MylesBartlett in #173

Full Changelog: v0.1.5...v0.1.6

v0.1.5

What's Changed

• Fix MeanStd by @thomkeh in #170
• Set max_pixel_value to 1.0 in Normalize by @thomkeh in #171

Full Changelog: 0.1.4...v0.1.5

0.1.4

What's Changed

• remove unneccessary dependency by @olliethomas in #168

Full Changelog: v0.1.3...0.1.4

v0.1.3

What's Changed

• release 0.1.2 by @MylesBartlett in #165
• progbar by @MylesBartlett in #166

Full Changelog: v0.1.2...v0.1.3

v0.1.2

What's Changed

• Bump version for new release by @thomkeh in #151
• Use new reduce_scatter function for groupwise metrics instead of broadcasting. by @MylesBartlett in #152
• release 0.1.1 by @MylesBartlett in #153
• Release script: auto-merge the PR by @thomkeh in #154
• bug/fix label mapping by @m4gpi in #155
• Fix: allow [m, n] labels over [n,] labels by @m4gpi in #148
• poetry update by @olliethomas in #156
• MeanStd not in all, but used by external libraries by @olliethomas in #157
• Update SSL losses. by @MylesBartlett in #159
• Remove cap on ranzen version. by @MylesBartlett in #160
• Cast collated samples to elem_type to avoid demotion in cdt_collate by @MylesBartlett in #161
• Add PACS dataset and datamodule. by @MylesBartlett in #162
• Construct path to subdirs using str(enum) instead of enum.value. by @MylesBartlett in #163
• Stop relying on the now-removed dims attribute in the base LightningDataModule class, replacing it with a conduit-defined dim_x attribute. by @MylesBartlett in #164

Full Changelog: v0.1.0...v0.1.2

v0.1.1

What's Changed

• Bump version for new release by @thomkeh in #151
• Use new reduce_scatter function for groupwise metrics instead of broadcasting. by @MylesBartlett in #152

Full Changelog: v0.1.0...v0.1.1

v0.1.0

What's Changed

• Test metric by @olliethomas in #1
• move mypy.ini to pyproject, and use master brach of lightning-bolts by @olliethomas in #2
• Eval cb by @olliethomas in #3
• Dann by @olliethomas in #4
• make losses > binary where possible by @olliethomas in #5
• test the inference steps by @olliethomas in #6
• Port OnlineReweightingLoss from real-patch. by @MylesBartlett in #7
• k*c instance weighting. by @olliethomas in #8
• Restructure codebase. by @MylesBartlett in #9
• add Zhang Gradient Project method by @olliethomas in #10
• Gpu tests by @olliethomas in #11
• Add ISIC dataset and dataset utility functions. by @MylesBartlett in #12
• Add NICO dataset with basic data-loading/downloading functionality (no biasing schemes yet). by @MylesBartlett in #13
• Add NICO preprocessing script by @MylesBartlett in #15
• Ph by @olliethomas in #14
• use a specified classifier by @olliethomas in #16
• Add a version of CelebA without the wrapper by @thomkeh in #17
• give option to not copy every callback to the inner trainer by @olliethomas in #18
• Fix problems with the new CelebA class that were discovered during testing by @thomkeh in #19
• Add split method to NICO that allows for concept/class conditional train/test splits. by @MylesBartlett in #20
• add parsable to datamodules by @olliethomas in #22
• Csa by @olliethomas in #21
• Add NICO data-module. by @MylesBartlett in #23
• Types by @olliethomas in #24
• losses are the wrong way around in gpd by @olliethomas in #25
• Rework datasets and data-modules. by @MylesBartlett in #26
• Fix preprocessing script for NICO dataset. by @MylesBartlett in #29
• SSRP dataset and str -> enum conversion in datasets by @MylesBartlett in #30
• Force kwarg-only where appropriate. by @MylesBartlett in #31
• add more EM datasets by @olliethomas in #32
• Use unified xent loss module from palkit. by @MylesBartlett in #33
• add s/y dim to dm by @olliethomas in #34
• fair mixup model by @olliethomas in #28
• rename dims to cardinality by @olliethomas in #35
• make underlying datatuple accessible by @olliethomas in #36
• add feature groups by @olliethomas in #37
• use EM for feature groups by @olliethomas in #38
• law by @olliethomas in #39
• return disc_features as a list of str, not feature groups by @olliethomas in #41
• Convert dataclasses to dictionaries in pb_default_collate instead of using astuple to avoid deep-copying. by @MylesBartlett in #43
• Add a base model class to reduce code-duplication. by @MylesBartlett in #40
• Allow val/test batch-size to be controllable independently of train batch-size. by @MylesBartlett in #44
• Add 'num_{split}_samples' properties to DMs. by @MylesBartlett in #45
• replace stage literal with enum by @olliethomas in #46
• Add 'promotion' methods to Sample classes. by @olliethomas in #47
• add enum str method to stop calling .value everywhere by @olliethomas in #48
• Initial audio dataset incorporation by @DavidHurst in #49
• Audio dataset base class implementation by @DavidHurst in #51
• fix some minor things by @olliethomas in #52
• Add tabular transforms and base tabular dataset. by @MylesBartlett in #50
• Ecoacoustics preprocessing and data downloading by @DavidHurst in #55
• Add Self-supervised methods. by @MylesBartlett in #54
• Use jar to unzip weird files by @olliethomas in #57
• conduit by @MylesBartlett in #58
• Split models into models and architectures. by @MylesBartlett in #59
• Update vit import in dino.eval. by @MylesBartlett in #60
• Mostly Linter Things by @olliethomas in #61
• Pad spectrograms terminally with zeros if not integer-divisible by specgram_segment_len by @MylesBartlett in #63
• Avoid str methods on path by @olliethomas in #62
• black by @olliethomas in #66
• Transform waveforms to spectrograms after segmenting. by @MylesBartlett in #67
• Add subset-factory function for CdtDataset instances. by @MylesBartlett in #70
• Add Waterbirds dataset. by @MylesBartlett in #72
• Update credit.py by @olliethomas in #74
• Add Waterbirds datamodule. by @MylesBartlett in #73
• Fix typos in comments and add docstrings. by @MylesBartlett in #75
• Add size property to base DM class. by @MylesBartlett in #76
• Override dims and not size in base DM class. by @MylesBartlett in #77
• Change CI so that it hopefully caches now by @thomkeh in #78
• Improve robustness of compute_instance_weights by using inverse indexes. by @MylesBartlett in #79
• make explicit that the transform should be passed as an arg by @olliethomas in #69
• add dummy tabular datamodule for testing by @olliethomas in #80
• Convert datamodule classes into attrs classes. by @MylesBartlett in #81
• remove lines from CI that load a cached poetry installation by @olliethomas in #83
• Add configen scripts and configen-generated files. by @MylesBartlett in #82
• Fix typo in warning message. by @MylesBartlett in #86
• Get property name from call stack for error message by @thomkeh in #87
• Add Relay class for orchestrating runs. by @MylesBartlett in #88
• Implement datamodule for ecoacoustics dataset by @thomkeh in #89
• Remove a few unused imports by @thomkeh in #92
• Fix type-annotations in with_hydra signature. by @MylesBartlett in #93
• Merge with-jar extracting code with download_from_url and pin versions of torch-related dependencies. by @MylesBartlett in #94
• Update lock file. by @MylesBartlett in #95
• Re-apply fix that was lost somehow by @thomkeh in #97
• Run poetry update and make sure the slow tests actually pass by @olliethomas in #96
• kit -> ranzen. by @MylesBartlett in #101
• clean up by @MylesBartlett in #102
• Run tests once a week on main (as regression tests) by @thomkeh in #103
• more data by @olliethomas in #100
• Specify extras for ranzen. by @MylesBartlett in #105
• Run slow tests for weekly test by @thomkeh in #104
• Port MNIST colorizer from ethicml so that ethicml dependency can be made optional. by @MylesBartlett in #106
• Add pandas as a mandatory dependency. by @MylesBartlett in #107
• Add TypeAlias everywhere where appropriate by @thomkeh in #108
• Don't run the slow tests in the scheduled CI by @thomkeh in #110
• Add Dummy Image Data by @olliethomas in #109
• dm test by @ol...