4.1.0 (2023-07-23)
- 424 allow any np.intX as training target (63a8dba)
- compute unsupervised loss using numpy (49bd61b)
- custom loss using inplace operations (423f7c4)
- disable ansi (60ec6bf)
- feature importance not dependent from dataloader (5b19091)
- README patience to 10 (fd2c73a)
- replace std 0 by the mean or 1 if mean is 0 (ddf02da)
- try to disable parallel install (c4963ad)
- typo in pandas error (5ac5583)
- update gpg key in docker file gpu (709fcb1)
- upgrade the ressource size (fc59ea6)
- use numpy std with bessel correction and test (3adaf4c)
- add augmentations inside the fit method (6d0485f)
- add warm_start matching scikit-learn (d725101)
- added conda install option (ca14b76), closes #346
- disable tests in docker file gpu to save CI time (233f74e)
- enable feature grouping for attention mechanism (bcae5f4)
- enable torch 2.0 by relaxing poetry (bbd7a4e)
- pretraining matches paper (5adb804)
- raise error in case cat_dims and cat_idxs are incoherent (8c3b795)
- update python (dea62b4)
3.1.1 (2021-02-02)
- add preds_mapper to pretraining (76f2c85)
3.1.0 (2021-01-12)
- n_a not being used (7ae20c9)
- save and load preds_mapper (cab643b)
3.0.0 (2020-12-15)
- checknan allow string as targets (855befc)
- deactivate pin memory when device is cpu (bd0b96f)
- fixed docstring issues (d216fbf)
- load from cpu when saved on gpu (451bd86)
- add new default metrics (0fe5b72)
- enable self supervised pretraining (d4af838)
- mask-dependent loss (64052b0)
2.0.1 (2020-10-15)
- add check for evalset dim (ba09980)
- pin memory available for training only (28346c2)
- specify device (46a301f)
- torch.load map_location in Py36 fallback (63cb8c4)
2.0.0 (2020-10-13)
- 1000 lines only when env=CI (c557349)
- add map_location to torch load (c2b560e)
- load_model fallback to BytesIO for Py3.6 (55c09e5)
- add check nan and inf (d871406)
- add easy schedulers (0ae114f)
- adding callbacks and metrics (1e0daec)
- refacto models with metrics and callbacks (cc57d62)
- speedups (5a01359)
- TabNetMultiTaskClassifier (5764a43)
- update readme and notebooks (9cb38d2)
1.2.0 (2020-07-01)
- verbosity with schedulers (d6fbf90)
- add entmax as parameter (96c8a74)
- allow other optimizer parameters (16d92d5)
- allow weights sample for regression (d40b02f)
- save and load tabnet models (9d2d8ae)
- save params and easy loading (6e22393)
1.1.0 (2020-06-02)
- allow zero layer (e3b5a04)
- sort by cat_idx into embedding generator (9ab3ad5)
- update forest_example notebook (8092324)
- add multi output regression (ffd7c28)
- add num_workers and drop_last to fit parameters (313d074)
- remove mask computations from forward (44d1a47)
- speed boost and code simplification for GBN (1642909)
1.0.6 (2020-04-20)
1.0.5 (2020-03-13)
- remove dead code for plots (f96795f)
- switch to sparse matrix trick (98910bc)
1.0.4 (2020-02-28)
- allow smaller different nshared and nindependent (4b365a7)
- sparsemax on train and predict epoch (6f7c0e0)
1.0.3 (2020-02-07)
- map class predictions for XGB results (3747e2f)
- fix shared layers with independent batchnorm (5f0e43f)
1.0.2 (2020-02-03)
1.0.1 (2020-01-20)
- fixing Dockerfile for poetry 1.0 (6c5cdec)
- importance indexing fixed (a8382c3)
- local explain all batches (91461fb)
- regression gpu integration an typos (269b4c5)
- regression: fix scheduler (01e46b7)
- resolve timer issue and warnings (ecd2cd9)
- improve verbosity (8a2cd87)
1.0.0 (2019-12-03)
- deps: update dependency numpy to v1.17.3 (eff6555)
- deps: update dependency numpy to v1.17.4 (a80cf29)
- deps: update dependency torch to v1.3.1 (18ec79b)
- deps: update dependency tqdm to v4.37.0 (f8f04e7)
- deps: update dependency tqdm to v4.38.0 (0bf45d2)
- functional balanced version (fab7f16)
- remove torch warnings (index should be bool) (f5817cf)