0.4.0

What's Changed

• Adding participation in PyTorch Ecosystem to Readme by @simona-rc in #285
• 🧹 Integrate nb-clean by @SagiPolaczek in #287
• added ability to ignore kwargs in function string descriptor building logic, and solved an issue in run_multiprocessed in verbose=0 case by @YoelShoshan in #286
• perplexity impl + post collect func process fix by @mosheraboh in #288
• Make logging optional by @shatz01 in #289
• few fixes to enable pl2 - and limit pl version until all tests will … by @mosheraboh in #291
• Support for pl2 in simple mnist example by @shatz01 in #294
• multiprocessing improvements by @YoelShoshan in #296
• Shatz batch sampler by @shatz01 in #299
• fixed num_available_cores to work correctly in distributed setting in LSF (using cores affinity) by @YoelShoshan in #300
• cindex: make pred np.array to allow negation by @michalozeryflato in #298
• add backbone mvit by @liamhazan in #293
• added an op to replace elements of value X with value Y - supporting numpy array, torch tensors, lists, and strings by @YoelShoshan in #302
• Support Lightning >=2.0.0 and Pandas >=2.0.0 by @SagiPolaczek in #301
• make components more general by @mosheraboh in #305
• mypy typing fix by @SagiPolaczek in #304
• resnet 3d with first channels dim != 64 bugfix by @itaijj in #308
• Resnet3d bug fix by @itaijj in #309
• add a flag to metric to avoid multigpu collection by @mosheraboh in #311
• Fix ISIC's ResNet run (Poriya) by @SagiPolaczek in #312
• Prep for long train by @mosheraboh in #313
• small fix. more intuitive condition for main process run by @alex-golts in #315
• update resnet with pool option, added contrastive supervised loss by @shakedpe in #314
• new picai seg by @itaijj in #310
• Shaked fuse by @shakedpe in #317
• cleaning and reorganizing fuse core code on segmentation by @itaijj in #318
• Refactoring: .flake8 & .mypy.ini -> setup.cfg by @SagiPolaczek in #319
• multiple validation dataloaders in common lightning module by @mosheraboh in #321
• Remove pyproject.toml by @SagiPolaczek in #323
• add an option in CollateDefault to add fixed items to batch_dict by @mosheraboh in #324
• add weight to extract loss class by @mosheraboh in #325
• Shaked fuse by @shakedpe in #327
• OpFunc: input=None passes the entire sample_dict by @michalozeryflato in #326
• Support a better string representation for NDict by @SagiPolaczek in #328
• add ignore_index to auc_roc metric by @michalozeryflato in #329
• adding torch_utils with select method similar to numpy by @michalozeryflato in #331
• clearml logger + improved mnist starter by @shatz01 in #330
• Added __len__ to PipelineDefault by @SagiPolaczek in #333
• Bfloat16 support by @mosheraboh in #334
• add a method to add sample ids to a dataset by @michalozeryflato in #332
• OpSetRandomChoice by @mosheraboh in #335
• Make op_id a really optional argument in OpCast by @SagiPolaczek in #337
• pearson correlation metric by @alex-golts in #338
• Added readthedocs config for documentation by @rakesh9177 in #339
• Make NDict (hopefully) lighter to import by @SagiPolaczek in #342
• Support TSV files in file_io by @SagiPolaczek in #341
• Add OpReplaceAnyElements to ops_common by @sivanravidos in #340
• Crop padding in collate by @liamhazan in #343
• support pl module test step by @sivanravidos in #344
• Perplexity gpu mem optimization by @mosheraboh in #345
• adds count num sequences and tokens metric by @mosheraboh in #346
• shared memory utility helping to speedup reading from files significantly if enough RAM is available by @YoelShoshan in #347
• Crop pad fix by @IdoAmosIBM in #348
• Bump GitHub workflows' python version to 3.9 by @SagiPolaczek in #350
• add workaround to support fsdp by @mosheraboh in #353
• collate now supports minimal dimension + added more support in samples cacher + added interface validation utility by @YoelShoshan in #349
• MetricCountSeqAndTokens counst tokens in label (if exist) + improving epoch metrics print by @michalozeryflato in #352
• implement hdf5 dataset reader op by @mosheraboh in #356
• use scipy implementation for Pearson correlation by @alex-golts in #355
• Matching keys lengths and new metrics by @IdoAmosIBM in #358
• min_size_per_dim + changes for scalars inputs+outputs support by @YoelShoshan in #354
• new OpRandCrop for augmentation by @bensha6757 in #360
• Oai example by @liamhazan in #361
• fixed bug of specificity calculation by @ellabarkan in #359
• clearml offline mode by @mosheraboh in #363
• added new metrics for regression tasks by @IdoAmosIBM in #364
• to backward support packages importing metrics from a previous sub module by @YoelShoshan in #365
• move modular tokenizer to fuse by @mosheraboh in #366
• support custom configure_optimizers method and avoid from saving random model by default by @mosheraboh in #367
• avoid from gpu sync attempt when using single gpu by @mosheraboh in #369
• Integrate ModularTokenizerOp with Hugging Face remote 🤗 by @SagiPolaczek in #368
• multi test dataloaders by @mosheraboh in #370
• adding r2 metric by @mosheraboh in #371
• Add new tokens to tokenizer op by @matanninio in #373
• rmse score fix by @mosheraboh in #372
• advancing to scalars gen2 by @YoelShoshan in #374
• Tokenizer's default behavior on unknown token by @SagiPolaczek in #375
• Bug fix in ModularTokenizer.decode() when the input is torch.Tensor type by @SagiPolaczek in #376
• Bump python version for MAMMAL release by @SagiPolaczek in #377

New Contributors

• @shakedpe made their first contribution in #314
• @rakesh9177 made their first contribution in #339
• @sivanravidos made their first contribution in #340
• @IdoAmosIBM made their first contribution in #348
• @bensha6757 made their first contribution in #360
• @matanninio made their first contribution in #373

Full Changelog: 0.3.0...0.4.0

0.3.0

Alert: major revision

FuseMedML new version introduces a few minor backward compatibility breaks.
Including new installation instructions.

What's Changed

• Add vit isic by @smartdanny in #221
• remove protobuf from dependencies by @alex-golts in #225
• Added more descriptive error messages and hints for common usage mistakes by @YoelShoshan in #226
• support arbitrary (not sequential and even not ints) balanced class values by @mosheraboh in #227
• update DL readme according to JOSS reviewer comments by @alex-golts in #232
• Some changes from DTI task (Add doc' and some small features) by @SagiPolaczek in #233
• knight data download script by @alex-golts in #235
• raise an exception in case of str input/outputs by @smartdanny in #234
• Auc metric fix by @avihu111 in #237
• CICD fix by @mosheraboh in #239
• fix learning rate and added config options by @liamhazan in #238
• Ensemble of multi-modal survival models by @michalozeryflato in #231
• add downsample option to BatchSamplerDefault by @mosheraboh in #246
• Ehr transformer ICU example by @ellabarkan in #245
• Simple MNIST Example by @smartdanny in #240
• Ehr transformer example by @ellabarkan in #256
• a more configureable resnet3d implementation by @avihu111 in #257
• not strict mode in ProjectPatchesTokenizer by @avihu111 in #260
• allow non-int sample ids in DatasetDefault by @alex-golts in #259
• support not specifying number of channels in ModelMultiHead by @alex-golts in #244
• update JOSS paper citation and badge by @alex-golts in #263
• support shallow copy of pipelines by @michalozeryflato in #262
• fix bug with specific binary input format by @alex-golts in #264
• Ndict support hydra by @michalozeryflato in #265
• add props to dataset and pipelines + small fix to cacher by @michalozeryflato in #266
• optimized train time for a use case of small samples and large batch by @mosheraboh in #268
• add multi-label metric cladd and func by @egozi in #269
• torchvision version upper bound fix by @SagiPolaczek in #254
• Additional model - CrossAttentionTransformerEncoder by @SagiPolaczek in #251
• ddp_fix by @smartdanny in #270
• NDict optimization by @SagiPolaczek in #271
• remove examples. prefix by @alex-golts in #276
• make examples subpackage (fuse_examples) by @mosheraboh in #277
• Fixes for the Hello World notebook by @SagiPolaczek in #275
• Pl model save + export the entire dataset (all keys) to dataframe by @mosheraboh in #281

New Contributors

• @ellabarkan made their first contribution in #245

Full Changelog: 0.2.9...0.3.0

0.2.9

What's Changed

• Color aug fix by @avihu111 in #213
• feat: added support for logits in some classification metrics #215 by @afoncubierta in #216
• adding check for class attribute, fixes issue #217 by @agiova in #218

Full Changelog: 0.2.8...0.2.9

0.2.8

What's Changed

• feat: add pre-commit hooks by @ishita1805 in #198
• Transformer + 1D/2D/3D Vision Transformer by @avihu111 in #200
• net changes from plotly visualization by @smartdanny in #202
• Add csv logger to some examples (mnist & stoic) by @smartdanny in #204
• ISIC using DDP strategy by @SagiPolaczek in #176
• Apply pre-commit changes (ex. mypy) on all files by @SagiPolaczek in #209
• loss wrapper to batch dict by @avihu111 in #206
• Update backbone_vit.py by @avihu111 in #207
• Ukbb changes by @avihu111 in #205
• fix: uncollate inference of batch size by @afoncubierta in #211

New Contributors

• @ishita1805 made their first contribution in #198
• @avihu111 made their first contribution in #200

Full Changelog: 0.2.7...0.2.8

0.2.7

What's Changed

• fix: explicitly exclude vulnerable protobuf version by @vagenas in #195

Full Changelog: 0.2.6...0.2.7

0.2.6

What's Changed

• fix(#192): relax protobuf constraint by @vagenas in #193
• added silent mode to evaluator by @smartdanny in #194

New Contributors

• @vagenas made their first contribution in #193
• @smartdanny made their first contribution in #194

Full Changelog: 0.2.5...0.2.6

0.2.5

What's Changed

• Lightweight mode in DatasetDefault which alllows good GPU utilization in large datasets (e.g. 100M samples) by @YoelShoshan in #177
• classification/regression Heads by @liamhazan in #164
• OpCond and knight sample ids by @mosheraboh in #191

Full Changelog: 0.2.4...0.2.5

0.2.4

What's Changed

• Add supported python versions badge by @SagiPolaczek in #183
• multimodal modifications by @mosheraboh in #184

Full Changelog: 0.2.3...0.2.4

0.2.3

What's Changed

• fix the case in dataset_default when sample_ids are integers by @simona-rc in #171
• fix (#172): support cpu in knight example by @agiova in #173
• Yoels by @YoelShoshan in #174
• Small documentation fixes by @SagiPolaczek in #175
• ML pipeline ("one click" cross validation) in fuse2 style by @alex-golts in #154
• Support Python 3.10 by @SagiPolaczek in #179
• Make ISIC multimodal again by @SagiPolaczek in #168
• PyPi update - bump version and classifiers by @SagiPolaczek in #182

New Contributors

• @simona-rc made their first contribution in #171
• @agiova made their first contribution in #173

Full Changelog: 0.2.2...0.2.3

0.2.2

What's Changed

• added remove_extension, changed default deepdiff to ignore nans, added proper implementation of keys() values() and items() to NDict, tried to also highlight more the faulting op in pipeline ops error by @YoelShoshan in #152
• Knight test by @liamhazan in #149
• override first backbone layers only when necessary by @mosheraboh in #156
• 🌐 Some small general (handle URLError, Ops Debugs, etc) by @SagiPolaczek in #157
• new wrapper by @itaijj in #160
• Improve knight by @liamhazan in #158
• move to BiomedSciAI org by @mosheraboh in #166
• remove opencv version limitation by @mosheraboh in #169

Full Changelog: 0.2.1...0.2.2