0615a80 | Anton Kiselev | 08 October 2019, 14:25:12 UTC | Fixes for serializing sets and numpy numbers in SerDe (#368) * Fixes for serializing sets and numpy numbers in SerDe * Tests for changes in SerDe. Fixes for JSON and code serializations | 08 October 2019, 14:25:12 UTC |
44d7b68 | Lucas Melin | 08 October 2019, 11:58:23 UTC | docs: fix typos in basic and extended forecasting tutorials (#370) | 08 October 2019, 11:58:23 UTC |
cea570d | guilherme-guilherme | 08 October 2019, 11:31:37 UTC | Embedding Input Vector (#315) * Option for DeepAR and DeepState to allow an embedding vector instead of the same value for all categorical features. By default, if no argument, automatically use the heuristic given by Jermey Howard of fast.Ai [1] Each categorical features with size cat_sz, will be associated with a vector with length defined as below. emb_szs = [(c, min(50, (c+1)//2)) for _, c in cat_sz] By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice. [1] https://github.com/fastai/fastai/blob/master/courses/dl1/lesson3-rossman.ipynb * https://github.com/awslabs/gluon-ts/pull/315 Ran "python setup.py type_check" - passed Ran "black src" for proper formatting * Ensuring embedding dimension is not None | 08 October 2019, 11:31:37 UTC |
40c20e3 | Lorenzo Stella | 08 October 2019, 11:12:07 UTC | add recursive batching/unbatching for list and tuple fields (#371) | 08 October 2019, 11:12:07 UTC |
8fb7e33 | Valentin Flunkert | 08 October 2019, 10:54:53 UTC | Revert "Started adding code-owners. (#351)" (#372) This reverts commit ab0befb3503cc476864a98337ff59184290214b2. | 08 October 2019, 10:54:53 UTC |
059e34f | Anton Kiselev | 03 October 2019, 12:27:52 UTC | Using custom SerDe methods for deserializing params in Sagemaker (#364) | 03 October 2019, 12:27:52 UTC |
af91b83 | Bo Chang | 01 October 2019, 21:28:31 UTC | Fix documentation (#362) | 01 October 2019, 21:28:31 UTC |
7e760a4 | Valentin Flunkert | 01 October 2019, 20:12:23 UTC | small fixes (#361) | 01 October 2019, 20:12:23 UTC |
5d141f6 | Iván Vallés Pérez | 01 October 2019, 19:43:28 UTC | Assure Learning Rate Scheduler does not increase the learning rate (#359) | 01 October 2019, 19:43:28 UTC |
6fdf7bd | Danielle Robinson | 01 October 2019, 18:53:42 UTC | Adding gpu support to the cholesky jitter and eig tests. (#342) * Adding gpu support to the cholesky jitter and eig tests. Also fixing the assert error for NaNs on the gpu * Simplifying assert since default sum and nansum return a scalar * Fix style check * Adding check for if the gpu is available we will run the tests on the gpu * Reconsolidating gpu check test with the checks in the core/component package | 01 October 2019, 18:53:42 UTC |
1ce8fe5 | Stephan Rabanser | 01 October 2019, 13:58:33 UTC | Improve dev setup instructions (#360) | 01 October 2019, 13:58:33 UTC |
c919005 | Jan Gasthaus | 01 October 2019, 13:39:50 UTC | Add synthetic data generation tutorial (#356) | 01 October 2019, 13:39:50 UTC |
0aa6e80 | Jasper Schulz | 30 September 2019, 19:41:09 UTC | Added pd.Timestamp to serde. (#357) | 30 September 2019, 19:41:09 UTC |
b7cf343 | Jan Gasthaus | 30 September 2019, 18:28:00 UTC | Recipe synthetic data generation small refactoring (#327) | 30 September 2019, 18:28:00 UTC |
695a226 | Jasper Schulz | 30 September 2019, 14:18:23 UTC | Actions run on pull-requst event. (#354) | 30 September 2019, 14:18:23 UTC |
ec06c79 | Jasper Schulz | 30 September 2019, 13:38:32 UTC | Marked test as flaky. (#353) | 30 September 2019, 13:38:32 UTC |
62bf164 | Valentin Flunkert | 30 September 2019, 13:20:57 UTC | small fix for binned distribution (#352) | 30 September 2019, 13:20:57 UTC |
ab0befb | Jasper Schulz | 30 September 2019, 10:58:16 UTC | Started adding code-owners. (#351) | 30 September 2019, 10:58:16 UTC |
f7ca507 | Valentin Flunkert | 30 September 2019, 09:42:57 UTC | small fixes to box cox (#349) | 30 September 2019, 09:42:57 UTC |
169ee70 | Valentin Flunkert | 30 September 2019, 09:40:24 UTC | Improve BinnedDistribution (#350) | 30 September 2019, 09:40:24 UTC |
22da305 | Valentin Flunkert | 30 September 2019, 08:08:26 UTC | Introduce ForecastGenerator to wrap mxnet output into forecast object (#348) * Introduce ForecastWrapper to wrap mxnet output into forecast object * rename ForecastWrapper -> ForecastGenerator | 30 September 2019, 08:08:26 UTC |
419fc44 | Danielle Robinson | 28 September 2019, 15:01:28 UTC | Adding GP example on synthetic dataset with built-in plotting (#343) * Adding GP example on synthetic dataset with built-in plotting * Fixing black | 28 September 2019, 15:01:28 UTC |
8f0291f | Jasper Schulz | 27 September 2019, 13:51:52 UTC | Increased test timeout. (#346) | 27 September 2019, 13:51:52 UTC |
6b89c33 | Jasper Schulz | 27 September 2019, 12:37:19 UTC | Adding tests through github action. (#308) | 27 September 2019, 12:37:19 UTC |
6992ce8 | Lorenzo Stella | 27 September 2019, 11:58:17 UTC | attempt to fix docs build (#345) | 27 September 2019, 11:58:17 UTC |
ad76eef | Lorenzo Stella | 27 September 2019, 11:12:08 UTC | add option for feat_static_real (#324) | 27 September 2019, 11:12:08 UTC |
e5ff6c5 | Jasper Schulz | 26 September 2019, 16:53:36 UTC | Added SageMaker batch transform support. (#317) * Added SageMaker batch transform support. | 26 September 2019, 16:53:36 UTC |
a97518d | Lorenzo Stella | 25 September 2019, 13:55:19 UTC | Fix DeepState samples tensor shape (#340) * fix deepstate samples tensor shape * address comments * address comment | 25 September 2019, 13:55:19 UTC |
29ab758 | Aaron Spieler | 20 September 2019, 13:24:27 UTC | switched to rollforward (#328) * switched to rollforward Timestamps will now be aligned using the rollforward method instead of rollback method. Additionally fixed incorrect comment. * Fixed tests related to rollforward switch. * Fixed formatting issues. | 20 September 2019, 13:24:27 UTC |
ae8239d | dcmaddix | 19 September 2019, 16:34:29 UTC | Adding FieldNames to artificial dataset module (#326) * Updating the artifical dataset module to use the defautl field names gluonts.dataset.field_names.FieldName as requested in PR 321 * Fixing style check * Removing code blocks that undid the prediction length chopping at end and context_length chopping at beginning | 19 September 2019, 16:34:29 UTC |
fd544ab | Lorenzo Stella | 19 September 2019, 14:53:00 UTC | fix a wrong reshaping (#330) | 19 September 2019, 14:53:00 UTC |
c0bc208 | Lorenzo Stella | 19 September 2019, 14:04:57 UTC | Add missing time features for business day frequency (#325) * add missing time features for business day freqency * move things around | 19 September 2019, 14:04:57 UTC |
515c45c | Valentin Flunkert | 16 September 2019, 21:40:40 UTC | Manage mxnet context when deserializing predictors (#318) | 16 September 2019, 21:40:40 UTC |
ff22722 | dcmaddix | 16 September 2019, 21:16:42 UTC | Changed item key to item_id in write_csv_row to reflect changes to the ArtificalDataset in PR #286 (#321) | 16 September 2019, 21:16:42 UTC |
1cac4c8 | Syama Rangapuram | 16 September 2019, 19:35:58 UTC | Assert that static cat features are used in case cardinality is explicitly set | 16 September 2019, 21:06:23 UTC |
f35929a | Valentin Flunkert | 16 September 2019, 20:56:22 UTC | include DistributionForecast and make plot generic (#316) | 16 September 2019, 20:56:22 UTC |
4479857 | Jan Gasthaus | 16 September 2019, 17:02:00 UTC | More operators for synthetic data generation (#286) Additional operators for generating synthetic data set, as well as minor changes to the synthetic data generation API. | 16 September 2019, 17:02:00 UTC |
96ea70d | Valentin Flunkert | 16 September 2019, 08:38:14 UTC | include quantile function in distribution (#314) | 16 September 2019, 08:38:14 UTC |
c2fbb47 | Valentin Flunkert | 15 September 2019, 20:58:45 UTC | make tensors and distributions serializable (#312) * make tensors and distributions serializable * make accuracy test for DeepFactor less strict | 15 September 2019, 20:58:45 UTC |
f20cf9e | Jasper Schulz | 12 September 2019, 16:13:48 UTC | Pytest-timeout==1.3; removes warnings from logs. (#306) | 12 September 2019, 16:13:48 UTC |
17897ce | Jasper Schulz | 12 September 2019, 16:13:30 UTC | Flask~=1.1; removes some warnings. (#307) | 12 September 2019, 16:13:30 UTC |
ce8c3e5 | Jasper Schulz | 12 September 2019, 15:22:01 UTC | Create black.yml (#305) | 12 September 2019, 15:22:01 UTC |
6ca5d5d | dcmaddix | 10 September 2019, 11:43:23 UTC | Added documentation and tests for Deep Factor https://github.com/awslabs/gluon-ts/issues/290 (#297) | 10 September 2019, 11:43:23 UTC |
cc7f36a | Thomas DENIMAL | 10 September 2019, 10:48:38 UTC | fixed bug when changing default activation function in wavenet #295 (#299) | 10 September 2019, 10:48:38 UTC |
48099c6 | Lorenzo Stella | 09 September 2019, 19:27:04 UTC | fix numerical issue with negative binomial distribution (#288) * fix numerical issue with negative binomial distribution * fix docstring * add missing header | 09 September 2019, 19:27:04 UTC |
2394ce5 | Konstantinos Benidis | 07 September 2019, 10:43:54 UTC | fixes fieldname issues (#292) * fixed fieldname and calc_stats | 07 September 2019, 10:43:54 UTC |
76f0d27 | Syama Sundar Rangapuram | 06 September 2019, 11:27:48 UTC | Bring deep state to gluon-ts (#229) * Add deep state space model * [Deepstate] address numerical stability (resulting from excess padding), add time features * [Deepstate] Add the missing headers * Addressing review comments * [Deepstate] correct typo in docstring * [Deepstate] Add reference for deep state in Bibliography; fix doc string | 06 September 2019, 11:27:48 UTC |
0c64a03 | Jasper Schulz | 05 September 2019, 13:00:41 UTC | Added Dockerfile. (#285) | 05 September 2019, 13:00:41 UTC |
fb041e4 | Jasper Schulz | 02 September 2019, 13:47:38 UTC | Fixed ProcessStartField. (#277) | 02 September 2019, 13:47:38 UTC |
4d044a2 | dcmaddix | 30 August 2019, 21:18:11 UTC | Deep Factor [WIP] (#271) * add deep factor * Modified shape documentation and test case for Deep Factor, so the metrics for electricity are approx 0.26 * add missing headers * Added __init__ file to DeepFactor to match those of other models | 30 August 2019, 21:18:11 UTC |
8721e92 | Lorenzo Stella | 30 August 2019, 09:41:42 UTC | fix pandas warnings in dataset generation (#270) * fix pandas warnings in dataset generation * use frequency_add | 30 August 2019, 09:41:42 UTC |
95201b0 | Lorenzo Stella | 29 August 2019, 21:02:40 UTC | fix mistake in notebook (#269) | 29 August 2019, 21:02:40 UTC |
ed3fa81 | Lorenzo Stella | 28 August 2019, 15:37:15 UTC | make item_id field uniform across predictors (#268) * make item_id field uniform across predictors * simplify code | 28 August 2019, 15:37:15 UTC |
aa579b9 | Jasper Schulz | 28 August 2019, 14:59:02 UTC | Update lag.py (#266) Updated error message. | 28 August 2019, 14:59:02 UTC |
e3fc99f | Jasper Schulz | 28 August 2019, 01:47:50 UTC | Removed `get_granularity`. (#265) Replaced with `pandas.tseries.frequencies.to_offset`. Relates to: #264 | 28 August 2019, 01:47:50 UTC |
7ef55e4 | Valentin Flunkert | 27 August 2019, 15:09:04 UTC | Improvements for wavenet and some utils (#262) - introduce train_window_length to make convolutions more efficient during training - make model hybridizable | 27 August 2019, 15:09:04 UTC |
4740da0 | Lorenzo Stella | 23 August 2019, 18:00:50 UTC | Fix shapes in CanonicalNetworkBase (#254) * harden test * fix scale shape * sort out tests better | 23 August 2019, 18:00:50 UTC |
4d710b0 | Lorenzo Stella | 23 August 2019, 15:04:18 UTC | update sphinx to 1.7.9 (#256) | 23 August 2019, 15:04:18 UTC |
ac8fc9b | Lorenzo Stella | 23 August 2019, 14:48:45 UTC | fix issue with MultivariateGaussianOutput (#257) | 23 August 2019, 14:48:45 UTC |
40e2de6 | Jasper Schulz | 23 August 2019, 02:16:25 UTC | Fix yearly freq in process start field. (#253) * Fix yearly freq in process start field. * Added rollback for business day freq. | 23 August 2019, 02:16:25 UTC |
657efde | dcmaddix | 22 August 2019, 04:32:07 UTC | Reformatted math equations for the log_prob method of the GaussianProcess class (#252) | 22 August 2019, 04:32:07 UTC |
6072fd9 | Alexander Alexandrov | 21 August 2019, 23:41:00 UTC | Fix documentation and enforce stricter doc builds (#226) - Remove Sphinx warnings and enable the `-W` Sphinx option in the docs build in order to avoid introduction of Sphinx warnings in the future. - Fix `__module__` for all `__init__` imports in order to resolve Sphinx errors caused by ambiguous cross-references. - Remove `docs/api/gluonts` from version control and synthesize them as part of the build process. - Generate one file in `docs/api/gluonts` per module. - Add a **References** section with bibliography references. - Add a menu link to the **Development Setup** section. - Remove orphaned **Quick Start** section. | 21 August 2019, 23:41:00 UTC |
261ee9f | Bernie Wang | 19 August 2019, 22:53:01 UTC | fix the shape error of the canonical network (#251) | 19 August 2019, 22:53:01 UTC |
9ff2fa1 | Bernie Wang | 16 August 2019, 21:35:07 UTC | fix the empty item list (#249) | 16 August 2019, 21:35:07 UTC |
83d0395 | Jasper Schulz | 16 August 2019, 21:13:23 UTC | Update __main__.py (#250) Fix error logging. | 16 August 2019, 21:13:23 UTC |
49e13b0 | Jasper Schulz | 16 August 2019, 00:55:41 UTC | Added error message to top of failure file. (#248) | 16 August 2019, 00:55:41 UTC |
47f272c | Jasper Schulz | 16 August 2019, 00:25:56 UTC | Adding failure information to failure file. (#247) | 16 August 2019, 00:25:56 UTC |
68bee6d | Jasper Schulz | 14 August 2019, 02:14:17 UTC | Black everything. (#244) | 14 August 2019, 02:14:17 UTC |
09dc692 | Jasper Schulz | 14 August 2019, 00:48:34 UTC | GluonTSException bypass fallback. | 14 August 2019, 01:07:06 UTC |
23e7a09 | Jasper Schulz | 13 August 2019, 22:47:25 UTC | Made `fallback` process-safe. Added ConstantValuePredictor. | 13 August 2019, 23:01:29 UTC |
18bd199 | Jasper Schulz | 13 August 2019, 19:14:53 UTC | Fix shell to have a canonical freq key in hyperparameters. | 13 August 2019, 19:50:56 UTC |
20aaa71 | Jasper Schulz | 13 August 2019, 19:13:57 UTC | Fix mean predictor. Forecasts always start at the end of the whole target. | 13 August 2019, 19:50:56 UTC |
99b0271 | Jasper Schulz | 13 August 2019, 19:12:29 UTC | Added fallback predictor and decorator. | 13 August 2019, 19:50:56 UTC |
b2d7517 | Jasper Schulz | 13 August 2019, 19:10:27 UTC | Added `predict_item` to RepresentablePredictor and adapted subclasses. (#240) * Added `predict_item` to RepresentablePredictor and adapted subclasses. Addresses #238. Also added a way to create a MeanPredictor based on an existing Predictor. | 13 August 2019, 19:10:27 UTC |
43ca1b7 | Jasper Schulz | 13 August 2019, 16:51:56 UTC | Adapted mean predictor to use random samples. (#239) * Adapted mean predictor to use random samples. | 13 August 2019, 16:51:56 UTC |
ffe17d5 | Gaurav Chauhan | 13 August 2019, 00:13:18 UTC | fix #234: Added method to fixup non json-spec compliant floats to make the resp… (#236) * Added method to fixup non json-spec compliant floats to make the response json compliant | 13 August 2019, 00:13:18 UTC |
3ad0216 | Jasper Schulz | 13 August 2019, 00:09:12 UTC | Moved simple predictors to a distinct model folder. (#237) | 13 August 2019, 00:09:12 UTC |
c93445a | Jasper Schulz | 12 August 2019, 23:27:27 UTC | Fixed backtest test. (#235) | 12 August 2019, 23:27:27 UTC |
e14ad69 | Konstantinos Benidis | 09 August 2019, 12:18:44 UTC | Fix num_eval_samples (#232) rename `num_eval_samples` to `num_parallel_samples` and fix some documentation | 09 August 2019, 12:18:44 UTC |
096b1cc | Jasper Schulz | 08 August 2019, 23:02:30 UTC | Removed logging for ping. (#233) | 08 August 2019, 23:02:30 UTC |
c0be26b | Lorenzo Stella | 07 August 2019, 21:25:04 UTC | Bump pandas version and remove timestamp workarounds (#230) * bump pandas version, remove workaround * removed workaround in npts as well * fix mistake | 07 August 2019, 21:25:03 UTC |
9a1fd93 | Jasper Schulz | 01 August 2019, 15:55:03 UTC | Added throughput logging for inference. (#228) | 01 August 2019, 15:55:03 UTC |
3d5b962 | Alexander Alexandrov | 26 July 2019, 15:01:41 UTC | Add utilities for shell tests Can be used to test SageMaker integration of different models. Demonstrated by the `test_shell` module that tests the shell itself. | 29 July 2019, 16:21:05 UTC |
3b34f5e | Alexander Alexandrov | 29 July 2019, 14:37:45 UTC | More flexible shell.serve API Introduces the following changes to the `serve` API in order to improve testability. - The `run_inference_server` is refactored as a factory method called `make_gunicorn_app` and now only constructs and returns the gunicorn instance but does not run it. - The `Settings` instance is now passed explicitly as a `make_gunicorn_app` parameter. - Add a `ServerFacade` wrapper class to `shell.serve`. Provides methods that abstract the REST API of a Gunicorn inference server. | 29 July 2019, 16:21:05 UTC |
432a85e | Alexander Alexandrov | 26 July 2019, 14:24:00 UTC | Add `MeanPredictor` to `model.testutil` Also change the `MeanEstimator` parameter from `num_samples` to `num_eval_samples` in alignment with the rest of the codebase. | 29 July 2019, 12:39:12 UTC |
9cce120 | Alexander Alexandrov | 25 July 2019, 14:22:59 UTC | Add support for artificial datasets in the dataset repository - Add support for `ArtificialDataset`-based datasets in the dataset repository through a new `generate_artificial_dataset` function. - Add `constant` as alias for the `ConstantDataset()`. | 25 July 2019, 17:07:19 UTC |
4f79513 | Alexander Alexandrov | 25 July 2019, 14:15:05 UTC | Minor changes in the `shell.sagemaker` package - Factor out `map_dct_values` as a `support.util` method. - Throw an assertion error if the `freq` key is not present after loading the hyperparamters dictionary. - Add encoding support for sagemaker hyperparamter dictionaries. - Pass `base` parameter as `Path` in `TrainPaths` and `ServePaths`. | 25 July 2019, 17:07:19 UTC |
e19bf0c | Alexander Alexandrov | 25 July 2019, 14:05:01 UTC | Improvements in the `core` package - Equality comparisons that recurse in dictionary and list values. - Support for encoding and equality comparisons of numpy arrays and data types. | 25 July 2019, 17:07:19 UTC |
93d7a56 | Alexander Alexandrov | 25 July 2019, 14:26:08 UTC | Serialize training metrics through the logger | 25 July 2019, 17:07:19 UTC |
ea8b748 | Alexander Alexandrov | 25 July 2019, 16:23:23 UTC | Add missing headers | 25 July 2019, 17:07:10 UTC |
ce262e1 | Alexander Alexandrov | 25 July 2019, 16:18:48 UTC | Add a script to check and format license headers | 25 July 2019, 17:07:10 UTC |
a2cc866 | nllosse | 23 July 2019, 11:52:55 UTC | Minor code style fixes (#220) * fix type hinting in backtest_metrics * replace assertion with regular Exception | 23 July 2019, 11:52:55 UTC |
a6d7e2b | Alexander Alexandrov | 22 July 2019, 17:08:45 UTC | Align MXNet version range in conda and docs (#217) | 22 July 2019, 17:08:45 UTC |
2527812 | Alexander Alexandrov | 22 July 2019, 12:08:44 UTC | Unify logging format in shell | 22 July 2019, 13:05:47 UTC |
d83e584 | Jasper Schulz | 22 July 2019, 09:23:07 UTC | Commit quantile output format. Now quantiles are a dict with quantile-name -> values. | 22 July 2019, 12:08:14 UTC |
1f499aa | Alexander Alexandrov | 22 July 2019, 10:04:08 UTC | Exclude MXNet 1.5.* from allowed requirements Seems that GluonTS has an issue with the new MXNet version. | 22 July 2019, 10:16:15 UTC |
b8be837 | Lorenzo Stella | 19 July 2019, 16:08:37 UTC | minor fix to installation docs (#210) | 19 July 2019, 16:08:37 UTC |
d05eaf6 | Konstantinos Benidis | 19 July 2019, 15:53:26 UTC | Transformer (#206) Added transformer model, tests and evaluations | 19 July 2019, 15:53:26 UTC |
df7b5bd | Alexander Alexandrov | 19 July 2019, 11:22:15 UTC | Bump up `setuptools_scm` version | 19 July 2019, 13:49:46 UTC |
35a3bbf | Bernie Wang | 19 July 2019, 01:07:52 UTC | add one more legacy frequency key (#208) | 19 July 2019, 01:07:52 UTC |