aa04074 | Louis de Gaste | 16 May 2023, 12:56:30 UTC | saving wip | 16 May 2023, 12:56:30 UTC |
6cfb108 | Louis de Gaste | 16 May 2023, 10:46:51 UTC | saving wip | 16 May 2023, 10:46:51 UTC |
41b1069 | Louis de Gaste | 16 May 2023, 08:43:17 UTC | saving wip | 16 May 2023, 08:43:17 UTC |
85499a9 | Louis de Gaste | 16 May 2023, 07:43:03 UTC | saving wip | 16 May 2023, 07:43:03 UTC |
0077ba7 | Louis de Gaste | 10 May 2023, 15:00:04 UTC | added prophet hyper parameters plus other paramspace tuning | 10 May 2023, 15:00:04 UTC |
9357b2f | Louis de Gaste | 09 May 2023, 08:54:12 UTC | typo fixing | 09 May 2023, 08:54:12 UTC |
40708cb | Louis de Gaste | 09 May 2023, 07:03:58 UTC | tuning and validation of the parameter space + adding the tools to visualise the output of the pipeline | 09 May 2023, 07:03:58 UTC |
a8a8593 | Louis de Gaste | 03 May 2023, 08:43:42 UTC | cleaned up the bench-marking process and refined the hyper-paramer space | 03 May 2023, 08:43:42 UTC |
94c18d0 | Louis de Gaste | 25 April 2023, 14:40:07 UTC | Merge branch 'docs/benchamrk-takeover' into docs/benchmarks # especially if it merges an updated upstream into a topic branche | 25 April 2023, 14:40:07 UTC |
d8d3249 | Louis de Gaste | 25 April 2023, 13:00:26 UTC | Fixed parameter incompatibility issues in parameter space search | 25 April 2023, 13:00:26 UTC |
7314300 | Louis de Gaste | 21 April 2023, 13:55:34 UTC | Merge branch 'master' into docs/benchmarks | 21 April 2023, 13:55:34 UTC |
08a7a54 | Louis de Gaste | 21 April 2023, 13:52:04 UTC | Merge branch 'master' into docs/benchamrk-takeover | 21 April 2023, 13:52:04 UTC |
1486e87 | Louis de Gaste | 21 April 2023, 13:50:20 UTC | 1 doc string | 21 April 2023, 13:50:20 UTC |
0db3464 | Louis de Gaste | 21 April 2023, 12:18:12 UTC | bug fix | 21 April 2023, 12:18:12 UTC |
4bbbdbc | Louis de Gaste | 19 April 2023, 13:02:16 UTC | turned optuna search to functionning state | 19 April 2023, 13:02:16 UTC |
7fd1cae | Louis de Gaste | 19 April 2023, 08:06:20 UTC | testing flake8 check for param_space.py | 19 April 2023, 08:06:20 UTC |
79ce54e | Louis de Gaste | 19 April 2023, 07:38:05 UTC | some file splitting | 19 April 2023, 07:38:05 UTC |
31989f9 | Louis de Gaste | 18 April 2023, 06:55:07 UTC | Taking over the benvchamrking task from Eliane. Currewntly trying to seperate code for optuna search/model evaluation/reference benchmark | 18 April 2023, 06:55:07 UTC |
f9dcfe9 | Leon Lowitzki | 13 April 2023, 18:55:10 UTC | Update torch_forecasting_model.py (#1708) | 13 April 2023, 18:55:10 UTC |
d3e9472 | dennisbader | 12 April 2023, 13:24:40 UTC | Release 0.24.0 | 12 April 2023, 13:24:40 UTC |
4c9ae45 | Dennis Bader | 12 April 2023, 12:05:25 UTC | update release workflow with new secrets | 12 April 2023, 12:05:25 UTC |
3c0ca3d | Dennis Bader | 12 April 2023, 07:31:45 UTC | Release 0.24.0 (#1702) * bump u8darts 0.23.1 * update changelog * update changelog * update changelog | 12 April 2023, 07:31:45 UTC |
106e1dc | Louis de Gaste | 12 April 2023, 06:52:47 UTC | Merge branch 'master' into docs/benchmarks | 12 April 2023, 06:52:47 UTC |
65861d1 | madtoinou | 11 April 2023, 22:15:49 UTC | Feat/lagged features names (#1679) * feat: create and store the lagged features names in the regression models * feat: adding corresponding tests in tabularization * fix: support any kind of Sequence to generate the lagged features name * feat: verify that the number of lagged feature names matches the feature_importances in the relevant regression models * fix: if any of the variate is a sequence of ts with different components names, create generic name for the corresponding variate, updated the tests * fix: using the same naming convention for the lagged components as the explainability module * refactor and fix some type hint warnings * simplified lagged feature name generation and moved out of regression model * fix regr model tests * fix create lagged data tests * fix small bug in unit test * fix bug in unittest from last PR --------- Co-authored-by: dennisbader <dennis.bader@gmx.ch> | 11 April 2023, 22:15:49 UTC |
317f092 | Dennis Bader | 11 April 2023, 14:10:00 UTC | Feat/model use static cov (#1700) * adapt forecasting model to only optionally use static covariates * adapt forecasting models to consider static covs or not * add support for optional static covs to TFTModels * added unittests for TFMs with optional static covs support * add static cov option to all regression models * adapt reg model add_static_covariates to enforce same dimensions * add unit tests * update add_static_covariates docs * add static covariates lagged data tests * adapt regression model static cov tests with new logic * minor docs update * call considers_static_covariates property instead of attribute | 11 April 2023, 14:10:00 UTC |
c497a03 | madtoinou | 11 April 2023, 09:25:50 UTC | Fix/dlinear padding for even kernel size (#1695) * fix: asymetrical padding when using even kernel size, moved the computation of the padding in the constructor instead of the forward method * feat: added unitest to cover both even and odd kernel size values --------- Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> | 11 April 2023, 09:25:50 UTC |
622ff06 | Dennis Bader | 10 April 2023, 18:53:58 UTC | possible fix for lint issue (#1698) * possible fix for lint issue * possible fix for lint issue part 2 | 10 April 2023, 18:53:58 UTC |
0401fe3 | Rajesh Balakrishnan | 10 April 2023, 18:26:38 UTC | [BUG-FIX] RegressionEnsembleModel : Error Furture covariates/Past Covariates… (#1660) * [BUG-FIX] RegressionEnsembleModel : Error when trying to predict with FutureCovariates | Suggested With Code Fix * linted the code added space after comma * simplified future covariate check for global ensemble it will check if supports past covariates. For both local and global model it will future covariates supports * test RegressionEnsemble Model with future Covariates Tests with local models supports future covariates along with models which don't support. Prophet() is used to represent one which support future covariate * Fixed possible lint issue Added blank line before comment * Lint correction with Black Linted code with black * isort fix manual fixed library import order * Replace Prophet with altenatives * simplify covariate usage --------- Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> Co-authored-by: dennisbader <dennis.bader@gmx.ch> | 10 April 2023, 18:26:38 UTC |
a480b5e | JanFidor | 10 April 2023, 17:08:29 UTC | Fix/ensemble historical forecasts (#1616) * add correct extreme_lags override and test * add required extreme_lags override * delete logging print * change lag priorities * add a test + use switch to tuple * fix extreme lags from other PR * make RegressionEnsembleModel work * small unit test fix --------- Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> | 10 April 2023, 17:08:29 UTC |
ebb9eb6 | madtoinou | 10 April 2023, 13:29:52 UTC | Fix/historical_forecasts callable retrain argument (#1675) * fix: better support for the Callable retrain argument in historical_forecast * fix: remove unused util function * fix: adding tests to cover historical forecast without lags but fut cov, and lags + fut cot + output_chunck_length > 3 * fix: updating the local model tests to run with the new constraints on the retrain argument of historical_forecasts * Apply suggestions from code review Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> * fix: addressing review comments * feat: testing retrain_func returning str and int * feat: adding utils method to get lags in the past only * fix: properly computing the timestamp shift for prediction, better support of situation where training is impossible, updated documentation about the start argument of historical_forecasts, create dummy support ts when using only future covariate (first timestamp of the input series is predictable) * feat: added unittest checking the start and end of historical_forecasts outputs for both retrain=True and retrain=False * feat: adressing review comments, better handling of the potential shift between the predictable and trainable timestamps, more detailed error messages * better handling of the TORCH_AVAILABLE variable (previously, env without torch were not running any historical_forecasts tests) * fix: better granularity in the unittests error catching * add some comments to hist forecasts test for interpretability * minor refactoring * refactor start handling in historical forecasts * fix residuals for models that do not require a minimum of two traning samples * fix an issue with new logic of retrain false at beginning * fix expected forecasting lengths for TFMs which require only 1 train sample * move from start error to warning * handle train length better * update documentation of hist fc, backtest, and gridsearch * revert to old sanity checks for start * improve start santiy check handling and error messages * fix small condition type in unit tests * move regression models out of flavor check in unit tests * make warnings default and use default start if outside of hist fc index * move start reset in historical forecasting --------- Co-authored-by: dennisbader <dennis.bader@gmx.ch> | 10 April 2023, 13:29:52 UTC |
e4908f7 | Jing Qiang Goh | 07 April 2023, 17:45:48 UTC | [Feature] Support a user defined function name in the window transformation output (#1676) * Support user-defined name in window transformation * Update doc string * Add tests * Review: Test respects the built-in function name --------- Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> | 07 April 2023, 17:45:48 UTC |
0307356 | Jing Qiang Goh | 07 April 2023, 17:45:20 UTC | [DOC] Improve the explanation on the window parameter (offset) of WindowTransformer (#1666) * Improve doc string to clarify the data type of offset expected by the window parameter of WindowTransformer Add remark to window_transformer file * Add test case to demonstrate the transform behaviors, considering different formats for the parameter window * Review: simply use self.assertRaises * Fix failed test --------- Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> | 07 April 2023, 17:45:20 UTC |
87b2b69 | Dennis Bader | 07 April 2023, 14:49:46 UTC | fixes sklearn Parallel issue for sklearn versions < 1.3 (#1697) | 07 April 2023, 14:49:46 UTC |
21c185d | Dennis Bader | 07 April 2023, 14:22:45 UTC | fixes lgb warning when not using any categorical features (#1681) * fixes lgb warning when not using any categorical features * fixes deprecation warning for sklearn Parallel delayed * fix failing unittest for categorical lgbm * Apply suggestions from code review Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> --------- Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> | 07 April 2023, 14:22:45 UTC |
e4a6933 | aarondt1 | 07 April 2023, 14:22:14 UTC | Fix/start value for historical_forecasts in examples (#1693) * Fixing the start value - Changing the start value in the text from 0.75 to 0.6 to match the code below. * fixing typo * Apply suggestions from code review --------- Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> | 07 April 2023, 14:22:14 UTC |
fcd12c2 | madtoinou | 07 April 2023, 13:33:07 UTC | [DOC] Improve the TFM user guide (#1661) * doc: reorganizing the TFM guide and extending the section about save/load as well as the callbacks * doc: adding command line instruction for tensorboard * doc: more detailed information about the NaNs in predictions * doc: creating a separate section for loading of weights, improved the table of content, added red crosses in covariate support tables * update tfm user guide * update tfm user guide * remove cloud storage section as it does not work --------- Co-authored-by: dennisbader <dennis.bader@gmx.ch> | 07 April 2023, 13:33:07 UTC |
69e83dc | Maxime Dumonal | 02 April 2023, 17:10:46 UTC | fix/historical_fore_fut_cov_notarget_lags (#1685) | 02 April 2023, 17:10:46 UTC |
5479108 | Levente Szabados | 30 March 2023, 09:46:18 UTC | implementing changes for exposing Lightning Tuner's LR finder (#1609) * implementing changes for exposing Lightning Tuner's LR finder * decouple dataset/loader generation from training * improve decoupling * add unit test and docs * add plotting from unittest * remove lr_finder example * fix minor mistakes in docs * final updates for PR * Update darts/tests/models/forecasting/test_torch_forecasting_model.py Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> * fix Tuner import for PTL<2.0.0 --------- Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> | 30 March 2023, 09:46:18 UTC |
bfbeb7c | Maxime Dumonal | 29 March 2023, 15:23:04 UTC | Fix/warning msg rnn (#1674) * warning * warning * adapt warning --------- Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> | 29 March 2023, 15:23:04 UTC |
ad944fc | Maxime Dumonal | 29 March 2023, 14:49:37 UTC | Fix/explain multimodel msg (#1673) * error msg + unittest add * corections * corrections * Update darts/explainability/shap_explainer.py Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> * cleaning unittest shap_explainer --------- Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> | 29 March 2023, 14:49:37 UTC |
de94ef4 | Rijk van der Meulen | 28 March 2023, 07:55:32 UTC | Functionality to let LightGBM effectively handle categorical features (#1585) * #1580 exploration * #1580 added cat_components to TimeSeries * #1580 _fit_model method LightGBM * #1580 included static covs in dummy unit test * #1580 integration with lgbm * #1580 helper func to method in RegressionModel * #1580 different approach; pass categorical covs to fit method of lgbm directly instead of attributes TimeSeries object * #1580 added few unit tests * #1580 small stuff * #1580 move categorical covs to model constructor * #1580 avoid code duplication in unit tests * #1580 add unit test on forecast quality with cat covs * #1580 add column names check in _get_categorical_covs helper * #1580 docstrings lgbm * #1580 add changelog entry * #1580 change check if ts has static cov * #1580 implemented RegressionModelWithCategoricalCovariates class * #1580 delete redundant test * #1580 replace test_quality_forecast_with_categorical_covariates unit test by suggestion Dennis * #1580 adjustment error messages validation method * #1580 adding categorical feature support for CatBoost * #1580 remove cat support CatBoost and smaller comments Dennis * #1580 finalizing * #1580 use parent _fit_model method * avoid creating lagged data twice * remove empty lines --------- Co-authored-by: Rijk van der Meulen <rijk.vandermeulen@eyeon.nl> Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> | 28 March 2023, 07:55:32 UTC |
28d3e2a | DavidKleindienst | 27 March 2023, 07:53:05 UTC | Fix prediction component name and add assertion to test (#1602) Co-authored-by: David Kleindienst <kleindienst@ximes.com> Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> | 27 March 2023, 07:53:05 UTC |
ba07326 | JanFidor | 26 March 2023, 19:06:23 UTC | Refactor forecasting models (#1590) | 26 March 2023, 19:06:23 UTC |
13b84c8 | Dennis Bader | 25 March 2023, 16:06:37 UTC | enfore multivariate series for VARIMA (#1657) | 25 March 2023, 16:06:37 UTC |
be21ed8 | JanFidor | 22 March 2023, 11:41:43 UTC | add already existing forecasts param (#1597) | 22 March 2023, 11:41:43 UTC |
e9ce199 | Dennis Bader | 21 March 2023, 19:54:08 UTC | avoid no checkpoint warning when loading from checkpoint (#1658) | 21 March 2023, 19:54:08 UTC |
537ffce | Maxime Dumonal | 21 March 2023, 10:56:17 UTC | fix/historical_forecasting_tft_futcov (#1652) * Correction in _get_historical_forecastable_time_index + unittests * Corrections unittests --------- Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> | 21 March 2023, 10:56:17 UTC |
1254ff6 | DavidKleindienst | 21 March 2023, 08:27:56 UTC | Make type hint python 3.7 compatible (#1656) Co-authored-by: David Kleindienst <kleindienst@ximes.com> | 21 March 2023, 08:27:56 UTC |
bb5b769 | DavidKleindienst | 20 March 2023, 19:17:12 UTC | Fix/Allow TimeSeries.with_hierarchy to accept strings as parents (#1584) * Convert hierarchy parents to list if string * Add comment Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> * Avoid mutating original hierarchy dict Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> * Test for equivalence of list&string hierarchy * Fix linting --------- Co-authored-by: David Kleindienst <kleindienst@ximes.com> Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> | 20 March 2023, 19:17:12 UTC |
afc2c3c | Dennis Bader | 20 March 2023, 16:08:54 UTC | fix failing test when lightning version <2.0.0 (#1654) | 20 March 2023, 16:08:54 UTC |
8b2df75 | Dennis Bader | 20 March 2023, 09:57:51 UTC | Fix/ptl version 200 (#1651) | 20 March 2023, 09:57:51 UTC |
95f4d4f | Błażej Nowicki | 19 March 2023, 12:03:00 UTC | Reducing testing time (#1625) * Marking slow tests * Group torch tests --------- Co-authored-by: Blazej Nowicki <blazej.nowicki@unit8.co> | 19 March 2023, 12:03:00 UTC |
f7c292a | Rijk van der Meulen | 12 March 2023, 17:37:24 UTC | Feature/add feature values to explainability result (#1546) * store feature values and shap.Explanation object in ExplainabilityResult * accounted for is_multioutputregressor * #1545 added entry to CHANGELOG.md * #1545 update docstrings for correctness API reference docs * #1580 create ShapExplainabilityResult subclass and remove decorator * #1580 adjust unit tests to have dedicated with statement per assert + other small stuff * #1580 change asserts in unit test from ExplainabilityResult to ShapExplainabilityResult * #1580 test get_feature_values() against raw output shap * #1580 adjust docstring * #1580 fixing small stuff * #1580 added one assert to unit test * #1545 implement _query_explainability_result() helper to avoid code duplication --------- Co-authored-by: Rijk van der Meulen <rijk.vandermeulen@eyeon.nl> Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> | 12 March 2023, 17:37:24 UTC |
301122e | Dennis Bader | 11 March 2023, 12:39:44 UTC | warn user if suspicion of wrong index in from_group_dataframe() (#1628) * warn user if suspicion of wrong index in TimeSeries.from_group_dataframe() * imroved warning message --------- Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> | 11 March 2023, 12:39:44 UTC |
fdc5a31 | Dennis Bader | 09 March 2023, 10:38:42 UTC | fix failing example notebook from MovingAverage rename (#1637) * fix failing example notebook from MovingAverage rename * rename MovingAverage in README | 09 March 2023, 10:38:42 UTC |
3c585d4 | Jeroen Van Der Donckt | 09 March 2023, 07:57:01 UTC | remove cap from statsforecast requirement (#1633) | 09 March 2023, 07:57:01 UTC |
6e16594 | madtoinou | 08 March 2023, 08:39:18 UTC | Fix/literal import (#1630) * fix: import Literal from typing extension to maintain compatibility with python 3.7 * fix: import Literal in a more elegant way | 08 March 2023, 08:39:18 UTC |
b07aaa8 | JanFidor | 07 March 2023, 22:22:03 UTC | Add quantile loss metric (#1559) * first implementation of quantile loss * add quantile loss to metrics ___init__ and tests * refactor * rename pinball loss to quantile loss * black * use reduction to aggregate losses and update docs * black + isort * rollback to simple mean instead of reduction param * change overlooked copy-paste comment * black enter * docs changes * flake8 --------- Co-authored-by: Julien Herzen <julien@unit8.co> Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> | 07 March 2023, 22:22:03 UTC |
92b74db | JanFidor | 07 March 2023, 20:27:44 UTC | Add NaiveMovingAverage baseline model (#1557) * implement NaiveMovingAverage model * rename Moving Average Filter * black * add NaiveMovingAverage to test local * delete debug print * black * isort * fix __init__.py file import * rename window param and delete unnecessary check * fix MA test param * add docs changes and deterministic assertion * use more readable assert syntax * add loger to raise_if_not Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> * black (again) --------- Co-authored-by: Julien Herzen <julien@unit8.co> Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> | 07 March 2023, 20:27:44 UTC |
8972d2e | Matt Bilton | 07 March 2023, 14:51:42 UTC | Refactor/data_transformers (#1409) * Refactored data transformers classes. * Fixed failing data transformer tests. * Fixed minor bug in `test_diff.py` - `~ bool_var` should be `not bool_var`. * Added missing `params` arg to pipeline test mock method. * Added automatic `component_mask`ing of inputs/outputs. * Added tests for data transformer classes. * Refactored data transformers classes. * Fixed failing data transformer tests. * Fixed minor bug in `test_diff.py` - `~ bool_var` should be `not bool_var`. * Added missing `params` arg to pipeline test mock method. * Added automatic `component_mask`ing of inputs/outputs. * Added tests for data transformer classes. * Fixed bug when fewer timeseries specified than training timeseries. * Updated tests to check for 'fewer inputs than training series' behaviour. * Added `global_fit` option to `FittableDataTransformer`. * Refactored `StaticCovariatesTransformer`. * Added `global_fit` option to `BoxCox` and `Scaler` transforms. * Removed `test_window_transformer_iterator` test, since `_transformer_iterator` method unused. * Removed redundant `_*_iterator` methods of data transformers. * Added more data transformer documentation + made `component_mask` argument explicit. * `copy=False` in `apply_component_mask`. * Removed documentation references to `_*_iterators`. * Specified `statsforecast>=1.4,<1.5` to avoid dependency conflict. --------- Co-authored-by: Julien Herzen <julien@unit8.co> Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> | 07 March 2023, 14:51:42 UTC |
e56faca | Dennis Bader | 07 March 2023, 10:00:22 UTC | Fix/hist forecast non retrain error (#1619) * raise an error in historical_forecasts() when retrain=False and model was not fit before * add unit tests for retrain values * black * fix failing tests * remove torch skip from test that is not required | 07 March 2023, 10:00:22 UTC |
7abb872 | Dennis Bader | 05 March 2023, 15:14:50 UTC | cap statsforecasts dependencies until fixed (#1620) | 05 March 2023, 15:14:50 UTC |
d2ba591 | DavidKleindienst | 28 February 2023, 19:09:04 UTC | Fix/Make reconcilliation order independent (#1592) * unit test for component order independence * remove incorrect assumption from test * Fix get_projection_matrix * minor improvement --------- Co-authored-by: David Kleindienst <kleindienst@ximes.com> Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> | 28 February 2023, 19:09:04 UTC |
32ea9a1 | Julien Herzen | 28 February 2023, 17:10:48 UTC | add a docstring entry to diff transformer (#1591) Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> | 28 February 2023, 17:10:48 UTC |
803d851 | madtoinou | 28 February 2023, 15:25:52 UTC | fix: replacing lambda with named function to make model pickable (#1594) * fix: replacing lambda with named function to make model pickable * fix: issue was also occurring with the exponential de-trending function * fix: adding typing * fix: linting --------- Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> | 28 February 2023, 15:25:52 UTC |
c80049b | RayDanks | 26 February 2023, 21:38:42 UTC | Upgrade isort to version 5.11.5 (#1578) | 26 February 2023, 21:38:42 UTC |
4110401 | Dennis Bader | 26 February 2023, 19:46:05 UTC | Feat/timeseries integer index (#1527) | 26 February 2023, 19:46:05 UTC |
d6a742b | Dennis Bader | 26 February 2023, 09:26:36 UTC | add note for fixing pruning callback issue (#1575) Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> | 26 February 2023, 09:26:36 UTC |
f0ea7e5 | madtoinou | 21 February 2023, 14:38:57 UTC | Feat/improved training from ckpt (#1501) * feat: new function fit_from_checkpoint that load one chkpt from the mode, allows user to change the optimizer, scheduler or trainer and export the ckpt of this fine-tuned model into another folder. fine-tuning cannot be chained using this method (original model ckpt must be reloaded) * fix: improved the model saving to allow chaining of fine-tuning, better control over the logger, made the function static * feat: allow to save the checkpoint in the same folder (loaded checkpoint is likely to be overwritten if the model is trained with default parameters) * fix: ordered arguments in a more intuitive way * fix: saving model after updating all the parameters to facilitate the chain-fine tuning * feat: support for load_from_checkpoint kwargs, support for force_reset argument * feat: adding test for setup_finetuning * fix: fused the setup_finetuning and load_from_checkpoint methods, added dcostring, updated tests * fix: changed the API/approach, instead of trying to overwrite attributes of an existing model, rather load the weights into a new model (but not the other attributes such as the optimizer, trainer, ... * fix: convertion of hyper-parameters to list when checking compatibility between checkpoint and instantiated model * fix: skip the None attribute during the hp check * fix: removed unecessary attribute initialization * feat: pl_forecasting_module also save the train_sample in the checkpoints * fix: saving only shape instead of the sample itself * fix: restore the self.train_sample in TorchForecastingModel * fix: update fit_called attribute to enable inference without retraining * fix: the mock train_sample must be converted to tuple * fix: tweaked model parameters to improve convergence * fix: increased number of epochs to improve convergence/test stability * fix: addressing review comments; added load_weights method and corresponding tests, updated documentation * fix: changed default checkpoint path name for compatibility with Windows OS * feat: raise error if the checkpoint being loaded does not contain the train_sample_shape entry, to make the break more transparent to users * fix: saving model manually directly after laoding it from checkpoint will retrieve and copy the original .ckpt file to avoid unexpected behaviors * fix: use random_state to fix randomness in tests * fix: restore newlines * fix: casting dtype of PLModule before loading the weights * doc: model_name docstring and code were not consistent * doc: improve phrasing * Apply suggestions from code review Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> * fix: removed warning in saving about trainer/ckpt not being found, warning will be raised in the load() call if no weights can be loaded * fix: uniformised filename convention using '_' to separate hours, minutes and seconds, updated doc accordingly * fix: removed typo * Update darts/models/forecasting/torch_forecasting_model.py Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> * fix: more consistent use of the path argument during save and load --------- Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> | 21 February 2023, 14:38:57 UTC |
955e2b5 | Levente Szabados | 21 February 2023, 12:55:26 UTC | Fixing logging and errors blocking multi GPU trianing of Torch models (#1509) * added fix for multi GPU as per https://pytorch-lightning.readthedocs.io/en/stable/extensions/logging.html#automatic-logging * trying to add complete logging in case of distributed to avoid deadlock * fixing the logging on epoch end for multigpu training * Black fixes for formatting errors * Added description of multi GPU setup do User Guide. --------- Co-authored-by: Julien Herzen <julien@unit8.co> Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> | 21 February 2023, 12:55:26 UTC |
690b6f4 | Julien Herzen | 17 February 2023, 10:20:59 UTC | Update make_dists.sh | 17 February 2023, 10:20:59 UTC |
3a0f1f1 | madtoinou | 16 February 2023, 14:18:16 UTC | fix: restore Python 3.7 back-compatibility (#1573) * fix: reverting lineterminator to lineterminator for compatibility with python 3.7 * fix: reverting n_init default values for kmeans for compatibility with python 3.7 | 16 February 2023, 14:18:16 UTC |
f2efd00 | julien12234 | 16 February 2023, 13:42:42 UTC | fix quantile detector when low/high threshold are the same (#1553) * fix quantile detector when low/high threshold are the same * add test * Update darts/ad/detectors/threshold_detector.py Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> * change syntax * pre commit * pre commit --------- Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> Co-authored-by: Julien Herzen <julien@unit8.co> | 16 February 2023, 13:42:42 UTC |
388d640 | dependabot[bot] | 16 February 2023, 12:03:00 UTC | Bump ipython from 8.8.0 to 8.10.0 in /requirements (#1558) Bumps [ipython](https://github.com/ipython/ipython) from 8.8.0 to 8.10.0. - [Release notes](https://github.com/ipython/ipython/releases) - [Commits](https://github.com/ipython/ipython/compare/8.8.0...8.10.0) --- updated-dependencies: - dependency-name: ipython dependency-type: direct:production ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: Julien Herzen <julien@unit8.co> | 16 February 2023, 12:03:00 UTC |
8c1bb2a | madtoinou | 16 February 2023, 11:57:50 UTC | Fix/solve warnings in unittests (#1502) * fix: warning about deprecated CountryHolidays * fix: precision loss occurend in moment calculation due to catastrophic cancellation by avoiding ts full of nan when computing skew and kurtosis * fix: warning message about logging interval behind to small for logtensorboard * fix: set default value of n_init of Kmeans to auto in anticipation of sklearn 1.4 * fix: adjusted n_cluster value for KMeansScorer to remove warning messages * fix: tweaking kernels length_scale_bounds to improve convergence * feat: added possibility to save last checkpoint without having warning message if model is trained without validation set * doc: added docstring for the additional argument in TorchForecastingModel * fix: updated deprecated keyword line_terminator to lineterminator * fix: removing deprecated freq argument from pd.Timetamp constructor calls * fix: removed deprecated argument freq from Timestamp constructor to remove warning * fix: revert changes related to the save_only_last_checkpoint argument in TorchForecastingModel * fix: remove call to deprecated freq in Timestamp * fix: set the argument n_init to 10, otherwise a test was failing because the new default value is better (AUC ROC increases from 0.9851 to 0.9935) * fix: closing the matplotlib figures to remove warning --------- Co-authored-by: Julien Herzen <julien@unit8.co> | 16 February 2023, 11:57:50 UTC |
84e8313 | DavidKleindienst | 16 February 2023, 08:53:14 UTC | Feature: Add support for logistic growth to Prophet (#1419) * Add logistic growth to prophet * Add RangeIndex to typehint and docstring * Fix import order * Move cap & floor to __init__ * Update darts/models/forecasting/prophet_model.py * Fix linting & remove duplicat docstring --------- Co-authored-by: David Kleindienst <kleindienst@ximes.com> Co-authored-by: Julien Herzen <julien@unit8.co> Co-authored-by: Julien Herzen <j.herzen@gmail.com> | 16 February 2023, 08:53:14 UTC |
e1c8d34 | Boyd Biersteker | 10 February 2023, 15:32:00 UTC | Improvement/statsforecastets: make sf_ets probabilistic + add future_covariate support for sf_ets + add AutoTheta (#1476) * StatsForecastETS now is probabilistic in the same way as StatsForecastAutoARIMA * include future covariates in sf_ets * sf_ets with future_covariates works.. probably it is underestimating the uncertainty because it doesn't take into account the uncertainty of the coef esimation of the OLS * Create separate file for StatsForecast models and extract some functions. * Added AutoTheta from the StatsForecast package. * Deleted sf_auto_arima.py and sf_ets.py, because the code is now included in sf_models.py. * Update darts/models/forecasting/sf_models.py Co-authored-by: Julien Herzen <j.herzen@gmail.com> * Update darts/models/forecasting/sf_models.py Co-authored-by: Julien Herzen <j.herzen@gmail.com> * Moved all statsforecast models to their own .py file. Added some comments explaining the handling of future covariates by StatsForecastETS. Included StatsForecastTheta in the tests. Moved the utility functions that the statsforecast models share to a singly .py file. Added the CES model which is supposed to be probabilistic, but that doesn't work yet eventhough it is supposed to be included in statsforecast 1.4.0. Trying to figure out why it isn't working. Removed sf_models.py. * Beginning of test for fit on residuals for statsforecast ets. * - AutoCES not probablisitc anymore, because that is not yet released in statsforecast 1.4.0 - changed AutoETS to SFAutoETS - added models to the base tests - wrote two units tests for future covariates use for sf_ets * - AutoCES not probablisitc anymore, because that is not yet released in statsforecast 1.4.0 - changed AutoETS to SFAutoETS - added models to the base tests - wrote two units tests for future covariates use for sf_ets * Changed StatsForecastETS to StatsForecastAutoETS. --------- Co-authored-by: Julien Herzen <julien@unit8.co> Co-authored-by: Julien Herzen <j.herzen@gmail.com> | 10 February 2023, 15:32:00 UTC |
31528a4 | ET | 10 February 2023, 10:55:48 UTC | fix deprecated references to darts-intro.ipynb / fix a sentence in tft (#1526) * fix deprecated references to darts-intro.ipynb / fix a sentence in tft * tft fixes only * fix tft without auto changed pics * change code-ref to link to 00-quick-start notebook --------- Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> Co-authored-by: Julien Herzen <julien@unit8.co> | 10 February 2023, 10:55:48 UTC |
c138315 | julien12234 | 10 February 2023, 10:50:39 UTC | fix model_fit_kwargs by removing none elements (#1520) Co-authored-by: Dennis Bader <dennis.bader@gmx.ch> Co-authored-by: Julien Herzen <julien@unit8.co> | 10 February 2023, 10:50:39 UTC |
55db01e | ET | 10 February 2023, 08:51:33 UTC | fix wrong import and default where missing. unify format (backticks) (#1535) Co-authored-by: madtoinou <32447896+madtoinou@users.noreply.github.com> | 10 February 2023, 08:51:33 UTC |
6887ed5 | ET | 07 February 2023, 13:55:38 UTC | rename all save_modle/load_model to save/load in docu (#1525) | 07 February 2023, 13:55:38 UTC |
9eaad7d | eliane-maalouf | 30 January 2023, 22:05:42 UTC | added results for dataset ETTh1Dataset | 30 January 2023, 22:05:42 UTC |
71e09ac | eliane-maalouf | 30 January 2023, 16:08:49 UTC | corrected README.md | 30 January 2023, 16:08:49 UTC |
f6c0d39 | eliane-maalouf | 30 January 2023, 15:31:03 UTC | corrected fitting with val series | 30 January 2023, 15:31:03 UTC |
39876ae | eliane-maalouf | 30 January 2023, 14:49:43 UTC | corrected error in plots | 30 January 2023, 14:49:43 UTC |
5f0275e | eliane-maalouf | 30 January 2023, 14:17:31 UTC | added plotting notebook for results bar charts | 30 January 2023, 14:17:31 UTC |
a5317e1 | eliane-maalouf | 29 January 2023, 00:03:29 UTC | added models, added readme | 29 January 2023, 00:03:29 UTC |
0a64368 | eliane-maalouf | 28 January 2023, 19:05:02 UTC | corrected scripts + added argparser | 28 January 2023, 19:05:02 UTC |
3dcadf4 | eliane-maalouf | 27 January 2023, 22:37:40 UTC | Merge branch 'master' into docs/benchmarks | 27 January 2023, 22:37:40 UTC |
4e46e10 | eliane-maalouf | 27 January 2023, 22:37:34 UTC | added experiments scripts | 27 January 2023, 22:37:34 UTC |
33d8a33 | Julien Herzen | 26 January 2023, 13:24:59 UTC | Fix/backtest multiple series (#1517) * add a unit test capturing a bug * fix a bug in backtesting with multiple series | 26 January 2023, 13:24:59 UTC |
7753a6b | Julien Herzen | 24 January 2023, 10:05:51 UTC | Merge branch 'master' of github.com:unit8co/darts | 24 January 2023, 10:05:51 UTC |
2c516e1 | Julien Herzen | 24 January 2023, 10:05:01 UTC | fix lcm call in tests for Python 3.7 | 24 January 2023, 10:05:01 UTC |
ec4e2cf | madtoinou | 23 January 2023, 18:11:35 UTC | Fix/allow empty datetimeindexed ts (#1499) * fix: datetime index completeness check is run only for non-empty ts, * fix: slicing method should not return TimeSeries starting before its start argument * fix: added corresponding tests Co-authored-by: Julien Herzen <julien@unit8.co> | 23 January 2023, 18:11:35 UTC |
de9374b | eliane-maalouf | 23 January 2023, 11:47:57 UTC | Merge branch 'master' into docs/benchmarks | 23 January 2023, 11:47:57 UTC |
2b07b06 | Julien Herzen | 23 January 2023, 11:19:11 UTC | handle Literal import in tabularization.py | 23 January 2023, 11:19:11 UTC |
864d190 | Matt Bilton | 23 January 2023, 10:29:01 UTC | Refactorised tabularisation + Jupyter notebook w/ experiments. (#1399) * Refactorised tabularisation + Jupyter notebook w/ experiments. * Added 'moving window' method + refactored 'time intersection' method. * Refactoring/code simplification + bug fixes. * Added `is_training` flag. * Added tests + bug fixes. * More tests + debugging. * Fixed zero lag value not allowed bug + other debugging. * RegressionModel now calls `create_lagged_training_data` + passing tests. * ShapExplainer now uses `create_lagged_prediction_data` + minor test modifications. * Added further documentation, esp to tests. * Moved `_add_static_covariates` from `tabularization.py` to `regression_model.py`. * Static covariates refactorings. * typo fix in test_regression_models.py Co-authored-by: eliane-maalouf <112691612+eliane-maalouf@users.noreply.github.com> * Removed old `_create_lagged_data` and tests notebook. * Clarification about `check_inputs` in docstring. * Allow `lags_future_covariates` to be `> 0`, and enforce `lags_past_covariates < 0`. * Made `get_feature_times` private, now `_get_feature_times`. * Placed `for` loop back inside `create_lagged_data`; more info in docstrings; test `Sequence[TimeSeries]` inputs and stochasic inputs. * Fixed `bootstrap=True` in `test_regression_models.py`. * Added note about `np.split` in `regression_model.py`. * Fixed repeated static covariates width calculation. * Fixed `shap_explainer` bug. * Amended static covariates test so that values of appended static cov columns are explicitly checked. * Updated docstring error. Co-authored-by: Matthew Bilton <mabilton@pop-os.localdomain> Co-authored-by: eliane-maalouf <112691612+eliane-maalouf@users.noreply.github.com> Co-authored-by: Julien Herzen <julien@unit8.co> | 23 January 2023, 10:29:01 UTC |
f88c1a2 | eliane-maalouf | 22 January 2023, 17:32:36 UTC | Merge branch 'master' into docs/benchmarks | 22 January 2023, 17:32:36 UTC |
5483e2f | Dennis Bader | 20 January 2023, 12:36:09 UTC | fix NoneType error for encoder train index generation (#1497) | 20 January 2023, 12:36:09 UTC |
a82ef5e | Julien Herzen | 20 January 2023, 10:35:44 UTC | Fix/tensorboard (#1500) * add tensorboardX to torch reqs * Make some tests deterministic * solve an issue with checking precision | 20 January 2023, 10:35:44 UTC |