46e22a7 | Lorenzo Stella | 10 May 2020, 16:10:28 UTC | bound negbin scale above 1 (#814) | 10 May 2020, 16:10:28 UTC |
b42a85f | Stephan Rabanser | 07 May 2020, 22:09:03 UTC | Add logit normal distribution (#811) | 07 May 2020, 22:09:03 UTC |
41bd8a5 | Kashif Rasul | 07 May 2020, 14:59:53 UTC | Fix negative binomial's scaling (#719) | 07 May 2020, 14:59:53 UTC |
b376349 | Jasper | 07 May 2020, 08:49:41 UTC | Disable TQDM when running on SageMaker. (#810) | 07 May 2020, 08:49:41 UTC |
e20c233 | Lorenzo Stella | 05 May 2020, 19:58:01 UTC | Fix LSTNet observed indicator usage (#804) * add observed indicator to blocks inputs * copy test changes from #802 * reduce test cases * use observed indicator as loss weight | 05 May 2020, 19:58:01 UTC |
a96fed4 | Leonardo Araneda Freccero | 01 May 2020, 04:57:00 UTC | Run nightly test (#797) * testing to run build on github * Latest release fetched over Github api The latest release is fetched over the Github API and is then cleaned up and use as the branch parameter in the git clone command. Thereafter the system install gluonts * merged installing of dependencies WIP * debug * debugging * Hopefully now running on windows too * Debug * Fixed typo * fix bug * debug * Splitting tag fetching into two tasks * Changed names and fixed bug * debug * Changed name and setup to run on schedule * Test running on latest release available via pip Debug run * Renamed the workflow files to better reflect what they do * Fixed bug * Fix flask dependency for Unix test * Jobs are now scheduled to run nightly and not on push Co-authored-by: Jasper <schjaspe@amazon.de> Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> | 01 May 2020, 04:57:00 UTC |
05d23a0 | Kashif Rasul | 30 April 2020, 20:14:08 UTC | LSTNet fixes (#791) * initial LSTNet fixes We need to use a Conv2D with appropriate kernel size * some fixes * Update src/gluonts/model/lstnet/_estimator.py Co-Authored-By: Lorenzo Stella <lorenzostella@gmail.com> * fix dimensions * fix typos * another typo * typo * fix rnn length * relax the metric a bit * scaler with keepdims=True * fix comment * add check for rnn length * Update test/model/lstnet/test_lstnet.py Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> | 30 April 2020, 20:14:08 UTC |
a55ced0 | Lorenzo Stella | 30 April 2020, 18:51:15 UTC | Update workflows against nightly mxnet (#801) | 30 April 2020, 18:51:15 UTC |
53120d4 | Kashif Rasul | 30 April 2020, 13:35:38 UTC | Use forecast_start in RForecastPredictor (#798) * fix RForecastPredictor fix pandas Timestamp * use forecast_start helper | 30 April 2020, 13:35:38 UTC |
8e6823a | Leonardo Araneda Freccero | 29 April 2020, 11:21:35 UTC | Fixing timeout issue for Wavenet (#793) * Tuned hyperparameter num_batches_per_epoch for wavenet speedup Increases speed of wavenet by ~34% when testing accuracy * Increased the allowed accuracy due to flaky accuracy values * The test also runs for windows machines * Changed num_batches_per_epoch to 3 This was discussed [here](https://github.com/awslabs/gluon-ts/pull/793#discussion_r416680005) | 29 April 2020, 11:21:35 UTC |
ed06add | Lorenzo Stella | 27 April 2020, 14:04:29 UTC | Add lead_time to predictors and estimators, update LSTNet (#700) * add lead_time attribute, update lstnet * fix r predictor constructor and docstring * remove wrong edit * make args to super constructors kwargs Co-authored-by: Lorenzo Stella <stellalo@amazon.de> | 27 April 2020, 14:04:29 UTC |
81eae04 | Kashif Rasul | 27 April 2020, 11:31:19 UTC | Scaler for (N,T,C) and (N,C,T) tensors (#790) * Scaler for (N,T,C) and (N,C,T) tensors fixed the scaler to work on time first and time last shaped tensors. * better doc * fix doc a bit more * added test * use axis argument as per review. * fix typo * fix test * Update src/gluonts/block/scaler.py Co-Authored-By: Lorenzo Stella <lorenzostella@gmail.com> * put back minimum_scale * fix test * fix nops test Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> | 27 April 2020, 11:31:19 UTC |
deef507 | Lorenzo Stella | 25 April 2020, 10:14:16 UTC | move naive predictors to independent subpackages (#788) | 25 April 2020, 10:14:16 UTC |
c71c94a | Lorenzo Stella | 25 April 2020, 09:16:00 UTC | Make data loader test flaky (#789) | 25 April 2020, 09:16:00 UTC |
785a289 | Valentin Flunkert | 24 April 2020, 13:02:40 UTC | fix bug with iteration over datasets (#787) * fix bug with iteration over datasets * remove batch shuffling and factor out bounds logic | 24 April 2020, 13:02:40 UTC |
cf4e599 | Caner Turkmen | 23 April 2020, 20:47:55 UTC | add variable length functionality to DataLoaders (#780) | 23 April 2020, 20:47:55 UTC |
6cc6ed4 | Lorenzo Stella | 22 April 2020, 09:21:32 UTC | Skip WaveNet test on windows (#777) * make windows build green again * Update test_model.py * Relax threshold It’s hard to set meaningful thresholds without assumptions on the hardware that runs the tests, so we should keep these small enough not to fail too often * Update test_loader.py * fix black | 22 April 2020, 09:21:32 UTC |
99e7e33 | Syama Sundar Rangapuram | 21 April 2020, 20:28:18 UTC | Fix feature slicing in WavenetSampler (#770) Co-authored-by: Syama Rangapuram <rangapur@amazon.de> Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> | 21 April 2020, 20:28:18 UTC |
39836e3 | Aaron Spieler | 21 April 2020, 12:13:33 UTC | Added tutorial notebooks and removed custom docker-files. (#678) * Added tutorial notebooks and removed custom docker-files. * Updating dependencies. | 21 April 2020, 12:13:33 UTC |
3b3b3e0 | Jasper Schulz | 19 April 2020, 12:17:14 UTC | Fix pydanitc `create_model` usage. (#768) Fixes #767 The `model_name` argument was renamed to `__model_name` in [1367](https://github.com/samuelcolvin/pydantic/pull/1367). This change ensure that our code works with both future and past versions of pydantic. | 19 April 2020, 12:17:14 UTC |
0143054 | Aaron Spieler | 17 April 2020, 13:54:11 UTC | MP DataLoader Improvements (#742) * Dataset iterations are now cache aligned and switched to default Pool method. * Added caching support for FileDataset. * Refactoring. * Added missing documentation in train loader. * Added more return types. * mend * Fixed bug regarding num_batches_for_sampling. * Reverting back to modulo based segmentation for code readability. * Massively simplified worker_fn due to simplified logic of num_batches_for_schuffling * mend * User warning in case of mp but not caching. * Minor refactoring. * Smaller reformatting. * Updated doc. * Simplified segmenting, test fix. * Yield from improvement. * Dataset Coverage Test Explicit. * removed print * Removed unused import. * Disabling windows mp evaluation, lowering required JSonLine throughput. Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> | 17 April 2020, 13:54:11 UTC |
c1fd03d | Lorenzo Stella | 17 April 2020, 08:05:14 UTC | remove deprecated features (#766) Co-authored-by: Lorenzo Stella <stellalo@amazon.de> | 17 April 2020, 08:05:14 UTC |
6210eac | Lorenzo Stella | 16 April 2020, 10:30:45 UTC | fix license header, update workflows (#762) * fix license header, update workflows * rename workflow Co-authored-by: Lorenzo Stella <stellalo@amazon.de> | 16 April 2020, 10:30:45 UTC |
78e5bc8 | Lorenzo Stella | 16 April 2020, 07:10:48 UTC | fix stack function (#759) Co-authored-by: Lorenzo Stella <stellalo@amazon.de> | 16 April 2020, 07:10:48 UTC |
661ce55 | Lorenzo Stella | 15 April 2020, 17:30:29 UTC | relax numpy requirement (#757) Co-authored-by: Lorenzo Stella <stellalo@amazon.de> | 15 April 2020, 17:30:29 UTC |
928283d | Jasper Schulz | 09 April 2020, 13:35:09 UTC | Remove direct python-dateutil dependency. (#748) * Remove direct python-dateutil dependency. * Update requirements/requirements.txt | 09 April 2020, 13:35:09 UTC |
db8c7f9 | Aaron Spieler | 08 April 2020, 12:06:47 UTC | Added multiprocessing support for evaluation. (#741) * Added multiprocessing support for evaluation. Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> | 08 April 2020, 12:06:47 UTC |
875b9d8 | Valentin Flunkert | 08 April 2020, 11:08:10 UTC | fix setup for multiprocessing (#747) | 08 April 2020, 11:08:10 UTC |
f25e8b3 | Stephan Rabanser | 07 April 2020, 23:19:57 UTC | Categorical Distribution (#746) * Add categorical distribution * Add tests for categorical distribution * Style fixes * Fix inference test * Add docstring to categorical distribution * Update hybridize testing parameter | 07 April 2020, 23:19:57 UTC |
4541451 | yx1215 | 07 April 2020, 22:34:52 UTC | change the default value to a scaler instance (#745) Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> | 07 April 2020, 22:34:52 UTC |
1aff85c | Syama Sundar Rangapuram | 07 April 2020, 21:49:51 UTC | Refactor wavenet model (#743) * Refactor wavenet model * [Wavenet refactoring] more docstrings Co-authored-by: Syama Rangapuram <rangapur@amazon.de> | 07 April 2020, 21:49:51 UTC |
a822353 | Aaron Spieler | 03 April 2020, 18:25:01 UTC | Mp data loader updates (#739) * Added mp support for inference loader. * Some refactoring regarding how the worker data is handled. * Removing superfluous conversion. * Fixed MXNet 0 in shape with workaround. Smaller refactoring. * Catch exception when shared memory resources are already freed. | 03 April 2020, 18:25:01 UTC |
cfe2729 | Aaron Spieler | 02 April 2020, 09:39:30 UTC | Multiprocessing data loader. (#689) * Datasets use the class attributes of MPWorkerInfo to get information about their multiprocessing environment. * Datasets are replicated among workers (only Object reference though, not the physical dataset), this happens exactly one in the beginning of the training * Datasets are not cached by default (caching not implemented so far) * Data loading can now be done in a multiprocessing fashion by specifying the number of workers, this works for training set and validation set * Parallelisation for datasets: modulo based, i.e. every num_workerth ts will be assigned to the corresponding worker. // however, this does not guarantee that the batches will always be sampled from equidistant locations for training, since some workers could potentially be slower or faster * The data loaders return batches of transformed samples of batch_size in the requested context. The transformation is done according to the provided transformation. * Which exact batches and samples one gets is nondeterministic if num_workers > 0 * Missing features: Shuffling (beyond a single batch), Dataset caching, Correct documentation * Current bugs: no mp support for windows or the InferenceDataLoader in general due to pickling errors. Co-authored-by: Valentin Flunkert <vafl@users.noreply.github.com> | 02 April 2020, 09:39:30 UTC |
147186e | Caner Turkmen | 01 April 2020, 11:53:13 UTC | add label smoothing to binned distribution (#731) | 01 April 2020, 11:53:13 UTC |
2e20016 | Lorenzo Stella | 30 March 2020, 18:43:17 UTC | remove asserts on loc and scale (#734) Co-authored-by: Lorenzo Stella <stellalo@amazon.de> | 30 March 2020, 18:43:17 UTC |
ed847c9 | Jasper Schulz | 27 March 2020, 09:18:50 UTC | Moved model tests into their own folders. (#727) | 27 March 2020, 09:18:50 UTC |
1108ed6 | Valentin Flunkert | 26 March 2020, 15:11:27 UTC | fix binned distribution for mxnet 1.6 (#728) * fix binned distribution for mxnet 1.6 * Update src/gluonts/distribution/binned.py Co-Authored-By: Lorenzo Stella <lorenzostella@gmail.com> Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> | 26 March 2020, 15:11:27 UTC |
a345629 | Lorenzo Stella | 25 March 2020, 18:04:21 UTC | Add MAPE to evaluator (#725) * add mape to evaluator * add mape to test cases Co-authored-by: Lorenzo Stella <stellalo@amazon.de> | 25 March 2020, 18:04:21 UTC |
015a980 | Lorenzo Stella | 25 March 2020, 12:47:58 UTC | remove unnecessary temporary patches (#724) * remove unnecessary temporary patches * update conda? * nevermind Co-authored-by: Lorenzo Stella <stellalo@amazon.de> | 25 March 2020, 12:47:58 UTC |
bc15c4e | Stephan Rabanser | 24 March 2020, 09:26:51 UTC | M5 data loader (#716) * Extend cardinality data type * Add M5 data loader | 24 March 2020, 09:26:51 UTC |
1108148 | Ehsan M. Kermani | 24 March 2020, 07:24:13 UTC | Remove hidden _base from extended tutorial (#714) Co-authored-by: Ehsan M. Kermani <kerehsan@amazon.com> | 24 March 2020, 07:24:13 UTC |
0f8119d | Lorenzo Stella | 18 March 2020, 14:31:20 UTC | Upgrade to mxnet 1.6 (#709) * remove "none" array placeholder * bump mxnet, numpy, update docs and test envs * Update README.md Co-Authored-By: Sheng Zha <szha@users.noreply.github.com> * Update docs/install.rst Co-Authored-By: Sheng Zha <szha@users.noreply.github.com> * exclude problematic tests * fix mistake * Update test/support/test_jitter.py Co-Authored-By: Jasper Schulz <jasper.b.schulz@googlemail.com> * Update test/support/test_jitter.py Co-Authored-By: Jasper Schulz <jasper.b.schulz@googlemail.com> * Update test/support/test_jitter.py Co-Authored-By: Jasper Schulz <jasper.b.schulz@googlemail.com> * fix skips * Apply suggestions from code review Co-authored-by: Lorenzo Stella <stellalo@amazon.de> Co-authored-by: Sheng Zha <szha@users.noreply.github.com> Co-authored-by: Jasper Schulz <jasper.b.schulz@googlemail.com> | 18 March 2020, 14:31:20 UTC |
9622550 | Jasper Schulz | 16 March 2020, 09:50:09 UTC | Prevent data from be cached in DataLoader. (#707) | 16 March 2020, 09:50:09 UTC |
331104f | Valentin Flunkert | 12 March 2020, 12:15:54 UTC | Improve trainer handling of learning rate scheduling and logging (#701) Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> | 12 March 2020, 12:15:54 UTC |
d47db85 | Stephan Rabanser | 10 March 2020, 10:35:20 UTC | Add M5 template (#699) * Add M5 template * Small syntax fixes * Add link to M5 competition | 10 March 2020, 10:35:20 UTC |
52f7aac | Ehsan M. Kermani | 29 February 2020, 00:39:22 UTC | Fix incompatible date_range and matplotlib register in pandas v1.0 (#679) Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> | 29 February 2020, 00:39:22 UTC |
1cdfced | Ehsan M. Kermani | 29 February 2020, 00:13:13 UTC | Add LSTNet (#596) * Add lstnet Add initial test Make smoke test pass Fix reshape bug and parameters validation Add more tests Fix unregistered hybrid rnn layer Add docs LSTNet scaling support Scaling and doc Fix dtype * Address the review comments Cleanup * Support horizon and prediction_length * Add more tests * Fix seq pred case and address comments * Simplify ar highway | 29 February 2020, 00:13:13 UTC |
f6c12a4 | Valentin Flunkert | 26 February 2020, 19:40:54 UTC | fix data loader and include validation channel in test (#680) | 26 February 2020, 19:40:54 UTC |
0c892b7 | Aaron Spieler | 26 February 2020, 15:37:22 UTC | Fixing docker images and adding test for sagemaker_sdk. (#656) * Added tests to test the gluonts train code of the sagemaker_sdk * Fixing docker images and adding tests to cover some of their functionality. * Marked makefile tests as xfail. Co-Authored-By: Jasper Schulz <jasper.b.schulz@googlemail.com> | 26 February 2020, 15:37:22 UTC |
a9c9f7e | Aaron Spieler | 26 February 2020, 15:08:05 UTC | Introducing baseline models. Includes new Naive 2 model, new OWA metric, and a lot of refactoring. (#602) * Added native python implementation of the Naive2 model and the OWA metric. * Various refactoring related to this. Co-authored-by: Jasper Schulz <jasper.b.schulz@googlemail.com>, Lorenzo Stella <lorenzostella@gmail.com> | 26 February 2020, 15:08:04 UTC |
28fd9be | Valentin Flunkert | 26 February 2020, 11:04:36 UTC | log epochs in trainer (#676) | 26 February 2020, 11:04:36 UTC |
47751d0 | Aaron Spieler | 24 February 2020, 15:02:14 UTC | Fixing mkl dependency issue, and further improvements. (#661) | 24 February 2020, 15:02:14 UTC |
e598800 | Aaron Spieler | 21 February 2020, 15:22:24 UTC | Updating yaml dependencies, in particular adding numpy explicitly. (#658) | 21 February 2020, 15:22:24 UTC |
53976b5 | Aaron Spieler | 20 February 2020, 12:52:28 UTC | N beats bugfix. (#655) * Fixes issue 629. * Added block naming for n_beats. | 20 February 2020, 12:52:28 UTC |
8013024 | Lorenzo Stella | 19 February 2020, 14:32:36 UTC | fixing context when converting to symbol block predictor (#651) | 19 February 2020, 14:32:36 UTC |
c21df0e | Valentin Flunkert | 19 February 2020, 08:25:43 UTC | support slicing of distributions (#645) | 19 February 2020, 08:25:43 UTC |
c65509f | Jasper Schulz | 18 February 2020, 12:01:56 UTC | Refactored instance sampler. (#648) | 18 February 2020, 12:01:56 UTC |
48a1b10 | Lorenzo Stella | 17 February 2020, 22:00:33 UTC | fix the sign method of affine transformation (#613) | 17 February 2020, 22:00:33 UTC |
887ce26 | Lorenzo Stella | 17 February 2020, 08:56:44 UTC | fix forecast index bug, add test (#644) | 17 February 2020, 08:56:44 UTC |
bbf2081 | Lorenzo Stella | 15 February 2020, 11:17:01 UTC | Update bug_report.md (#643) | 15 February 2020, 11:17:01 UTC |
c44f0a0 | Lorenzo Stella | 14 February 2020, 13:37:45 UTC | Fix some mypy issues and add type annotations in forecast.py (#639) * fix some mypy issues and add type annotations * fix mistake | 14 February 2020, 13:37:45 UTC |
a43a7e7 | hiok2000 | 14 February 2020, 09:54:06 UTC | Fix r-forecast package on windows. (#626) | 14 February 2020, 09:54:06 UTC |
0886ce4 | Jasper Schulz | 14 February 2020, 08:24:10 UTC | Removed logger.basicConfig calls. (#635) | 14 February 2020, 08:24:10 UTC |
24af514 | Jasper Schulz | 13 February 2020, 14:35:07 UTC | Removed `dataset.constants.py`. (#601) | 13 February 2020, 14:35:07 UTC |
cddd983 | Jasper Schulz | 13 February 2020, 13:57:14 UTC | Refactored DataLoader. (#619) | 13 February 2020, 13:57:14 UTC |
bc24d1c | Bo Chang | 12 February 2020, 21:06:50 UTC | Update _base.py (#630) | 12 February 2020, 21:06:50 UTC |
311ce4e | hiok2000 | 12 February 2020, 17:40:46 UTC | Update identity.py (#627) Forecast data should start from current position. Co-authored-by: Jasper Schulz <jasper.b.schulz@googlemail.com> | 12 February 2020, 17:40:46 UTC |
90a35bf | Jasper Schulz | 12 February 2020, 16:23:08 UTC | Mypy update (#628) * Update mypy. | 12 February 2020, 16:23:08 UTC |
1677728 | Jasper Schulz | 12 February 2020, 13:31:26 UTC | Updated black. (#625) | 12 February 2020, 13:31:26 UTC |
357c863 | Lorenzo Stella | 12 February 2020, 12:21:41 UTC | fix mean_ts method (#624) * fix mean_ts method * address comments * address comments | 12 February 2020, 12:21:41 UTC |
90587ba | Aaron Spieler | 07 February 2020, 01:13:49 UTC | Fixed minor bugs in GluonTSFramework. (#585) * Couple of minor fixes in the GluonTSFramework. Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> and Jasper Schulz <schjaspe@amazon.de> | 07 February 2020, 01:13:49 UTC |
b6ec6a0 | Aaron Spieler | 05 February 2020, 12:59:22 UTC | More elegant stop_gradient fix for sMAPE. (#593) * stop_gradient in denominator of sMAPE | 05 February 2020, 12:59:22 UTC |
8a62e00 | Aaron Spieler | 31 January 2020, 13:45:06 UTC | Updated N-Beats defaults. (#588) * Updated N-Beats defaults. * Adjustied documentation wording. Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> | 31 January 2020, 13:45:06 UTC |
6f0c16a | Caner Turkmen | 31 January 2020, 13:37:25 UTC | fix for symbol block import backward compatibility (#591) | 31 January 2020, 13:37:24 UTC |
e28cb5b | Ehsan M. Kermani | 31 January 2020, 12:59:09 UTC | Change causal conv1d default activation to match the doc (#586) Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> | 31 January 2020, 12:59:09 UTC |
5d13a3e | mehdikchouk | 31 January 2020, 09:50:08 UTC | Update extended_tutorial.md (#590) * Update extended_tutorial.md * Update docs/examples/extended_forecasting_tutorial/extended_tutorial.md Co-Authored-By: Lorenzo Stella <lorenzostella@gmail.com> Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> | 31 January 2020, 09:50:08 UTC |
2889be4 | Caner Turkmen | 30 January 2020, 20:01:29 UTC | Fix for symbol block serialization issue (#582) | 30 January 2020, 20:01:29 UTC |
ae0be7b | Lorenzo Stella | 30 January 2020, 17:08:20 UTC | disabling test for ensemble nbeats because of timeouts (#580) * disabling test for ensemble nbeats because of timeouts * disabling more test cases | 30 January 2020, 17:08:20 UTC |
f542207 | Jasper Schulz | 30 January 2020, 16:01:34 UTC | Fix Dockerfile to use Python 3.7. (#579) Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> | 30 January 2020, 16:01:34 UTC |
42f512a | Lorenzo Stella | 30 January 2020, 15:27:21 UTC | fix trend model to work in symbolic mode (#578) * fix trend model to work in symbolic mode * explicitly enable hybridization | 30 January 2020, 15:27:21 UTC |
6693927 | Jasper Schulz | 30 January 2020, 14:17:40 UTC | Enhanced error message for input time-series fields. (#577) * Enhanced error message for input time-series fields. * Fixed notebooks to use right ts shapes. | 30 January 2020, 14:17:40 UTC |
8b7775f | Jasper Schulz | 30 January 2020, 13:32:51 UTC | Pandas v1.0 (#576) | 30 January 2020, 13:32:51 UTC |
95a9d78 | Jasper Schulz | 30 January 2020, 09:14:47 UTC | Removed expand_dims when reading in time-series values. (#574) | 30 January 2020, 09:14:47 UTC |
2f10256 | Lorenzo Stella | 29 January 2020, 16:14:52 UTC | clean up lifted operations, add pow operation (#571) | 29 January 2020, 16:14:52 UTC |
611a1fd | Jasper Schulz | 29 January 2020, 14:38:47 UTC | Moved gp module to be part of gp_forecaster. (#572) | 29 January 2020, 14:38:47 UTC |
f44304d | Aaron Spieler | 29 January 2020, 13:25:36 UTC | N-Beats implementation for GluonTS (#553) * Implemented N-Beats as described in https://arxiv.org/abs/1905.10437. * exception: L_H parameter (sampling training sequences is done through InstanceSplitter) Co-Authored-By: Lorenzo Stella <lorenzostella@gmail.com> | 29 January 2020, 13:25:35 UTC |
76bbc6c | Lorenzo Stella | 27 January 2020, 14:43:08 UTC | add link to master docs, add docs badges (#570) | 27 January 2020, 14:43:08 UTC |
79685c1 | Lorenzo Stella | 27 January 2020, 09:54:43 UTC | add missing shape info to scale (#569) | 27 January 2020, 09:54:43 UTC |
b056435 | Aaron Spieler | 22 January 2020, 09:29:56 UTC | Fixed docstring. (#568) | 22 January 2020, 09:29:56 UTC |
1171f9e | Lorenzo Stella | 21 January 2020, 10:02:22 UTC | fix item_id field in provided datasets (#566) | 21 January 2020, 10:02:22 UTC |
824159c | Kashif Rasul | 20 January 2020, 21:28:30 UTC | fixed taxi dataset cardinality (#552) * fixed taxi dataset cardinality * Update _gp_copula_2019.py Points now to the corrected taxi file. * Update _gp_copula_2019.py Fix style check * Update _gp_copula_2019.py Minor. * Update _gp_copula_2019.py Style fix Co-authored-by: Michael Bohlke-Schneider <mschneid83@gmail.com> | 20 January 2020, 21:28:29 UTC |
6338eb2 | Lorenzo Stella | 20 January 2020, 15:10:13 UTC | Fix bug in cdf method of piecewise linear distributions (#564) * fix bug in cdf method of piecewise linear distributions * improve test * regularize slopes and add robustness test | 20 January 2020, 15:10:13 UTC |
55ee968 | Jasper Schulz | 16 January 2020, 18:58:16 UTC | Removed timeouts from tests. Added hard timeout of 30seconds for all tests. (#562) | 16 January 2020, 18:58:16 UTC |
c58cf39 | Lorenzo Stella | 15 January 2020, 17:02:14 UTC | remove dataset class in favor of alias (#560) | 15 January 2020, 17:02:14 UTC |
d647aa4 | Lorenzo Stella | 10 January 2020, 07:00:02 UTC | prob -> log_prob (#556) | 10 January 2020, 07:00:02 UTC |
753add5 | Micky C | 07 January 2020, 23:17:53 UTC | add poisson distribution (#532) Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> | 07 January 2020, 23:17:53 UTC |
d5a52c7 | Aaron Spieler | 07 January 2020, 13:49:40 UTC | Distribution dependent loss masking. (#534) * Loss is now masked with F.where for unobserved values. * Value in support exposed through DistributionOutput. Co-authored-by: Lorenzo Stella <lorenzostella@gmail.com> | 07 January 2020, 13:49:40 UTC |
0bfa906 | Kashif Rasul | 07 January 2020, 12:38:47 UTC | Shopping holidays (#542) * shopping holidays added black friday and cyber monday * fix for #469 fixes #469 by adding 1/2 a week more days * formatting * use Day | 07 January 2020, 12:38:47 UTC |
a4e5584 | Lorenzo Stella | 06 January 2020, 16:53:32 UTC | Add `loc` argument to distribution output classes (#540) * add loc to distribution output classes * fix docstring | 06 January 2020, 16:53:32 UTC |
4bf7f52 | Valentin Flunkert | 02 January 2020, 10:25:15 UTC | fix quantile fn for transformed distribution (#544) | 02 January 2020, 10:25:15 UTC |