Backporting fixes:

• Add test cases for PandasDataset, fix missing assertion (#2453 by @lostella)
• Speed up PandasDataset further (#2441 by @lostella)
• Fix MANIFEST.in (#2456 by @lostella)

Backporting fixes:

• Add test cases for PandasDataset, fix missing assertion (#2453 by @lostella)
• Speed up PandasDataset further (#2441 by @lostella)
• Fix MANIFEST.in (#2456 by @lostella)

Backporting fixes:

• Fix rotbaum random seed and num_samples argument. (#2408 by @sighellan)
• Hierarchical: Make sure the input S matrix is of right dtype (#2415 by @rshyamsundar)
• Mypy fixes (#2427 by @jaheba)
• Speed up PandasDataset for long dataframes (#2435 by @lostella)
• Fix frequency inference in PandasDataset (#2442 by @lostella)
• Tests: Change Python versions. (#2448 by @jaheba)

Backporting fixes:

• Fix dominick dataset bug. (#2364 by @haskarb)
• Remove strange quoting marks from docstrings (#2368 by @lostella)
• Consistent use of term "prediction interval" (#2373 by @codingWhale13)
• Fix MQCMM ignoring zero-seed. (#2379 by @sighellan)

Backporting fixes:

• Fix numerical bug in BinnedUniforms (#2344 by @moudheus)
• Fix dominick dataset bug. (#2364 by @haskarb)
• Fix MQCMM ignoring zero-seed. (#2379 by @sighellan)

Overview

Incremental training

Estimators are now re-trainable on new data, using the train_from method. This accepts a previously trained model (predictor), and new data to train on, and can greatly reduce training time if combined with early stopping. The feature is integrated with gluonts.shell-based SageMaker containers, and can be used by specifying the additional model channel to point to the output of a previous training job. More info in #2249.

New models

Two models are added in this release:

• DeepVARHierarchicalEstimator, a hierarchical extension to DeepVAREstimator; learn more about how to use this in this tutorial.
• DeepNPTSEstimator, a global extension to NPTS, where sampling probabilities are learned from data; learn more on how to use this estimator here.

Deprecated import paths and options

This release moves MXNet-based models from gluonts.model to gluonts.mx.model; the old import paths continue working in this release, but are deprecated and will be removed in the next release. For example, now the MXNet-based DeepAREstimator should be imported from gluonts.mx (or gluonts.mx.model.deepar).

We also removed deprecated options for learning rate reduction in the gluonts.mx.Trainer class: these can now be controlled via the LearningRateReduction callback.

Dataset splitting functionality (experimental)

We updated the functionality to split time series datasets (along the time axis) for training/validation/test purposes. Now this functionality can be easily accessed via the split function (from gluonts.dataset.split import split); learn more about this here.

This feature is experimental and subject to future changes.

Changelog

Breaking changes

• Breaking: Update data splitters to return (input, output) pairs in the test split (#2031 by @npnv)
• Breaking: Move MXNet-based models to mx.model. (#2126 by @Hongqing-work)
• Convert time-features into functions. (#2149 by @jaheba)
• Remove deprecated args from mx.Trainer. (#2153 by @jaheba)
• Reduce sdist size. (#2199 by @jaheba)
• Remove core.exception module. (#2202 by @jaheba)
• Remove core.ty. (#2203 by @jaheba)
• Update gluonts.dataset.split code, test, docs (#2223 by @lostella)
• Remove gluonts_forecasters entrypoint mechanic. (#2278 by @jaheba)
• Enable 'python -m gluonts'. (#2292 by @jaheba)

New features / major improvements

• Interrupting mx.Trainer stops training. (#2131 by @Hongqing-work)
• Expose evaluator aggregation_strategy functions (#2198 by @kashif)
• Add data preparation utility for hierarchical time series and a tutorial notebook (#2206 by @rshyamsundar)
• Add Deep NPTS model (#1835 by @rshyamsundar)
• Improve arrow reading performance. (#2217 by @mr-1993)
• Allow DeepVAR model to use (global) dynamic features (#2226 by @rshyamsundar)
• Hierarchical: Allow use of external dynamic features and add a section in the tutorial (#2253 by @rshyamsundar)
• Add serde.dataclass. (#2166 by @jaheba)
• R: Add Python wrapper for calling R's hierarchical methods (#1685 by @rshyamsundar)
• Add learning rate and weight decay arguments to PyTorch estimators (#2289)
• Added LR scheduler to DeepAR Pytorch (#2287 by @shubhamkapoor)
• Add LR scheduling patience option to MQF2 (#2291 by @lostella)
• Add incremental training (#2249 by @lostella)
• Add input size and type information to DeepARModel, and example_input_array to DeepARLightningModule. (#2307 by @jgasthaus)
• Add dataset.schema.translate. (#2304 by @jaheba)
• Add forecast_start to entry-wise metrics in evaluator (#2312 by @lostella)

Bug fixes / minor improvements

• Fix DatasetCollection (#2135 by @rsnirwan)
• Fix PandasDataset for Python 3.9 (#2141 by @lostella)
• Make PandasDataset faster (#2148 by @lostella)
• Ignore divide warnings in evaluation. (#2159 by @jaheba)
• Fix Prophet wrapper to work with Timestamp instead of Period (#2182 by @lostella)
• Fix dtype for "item_id" column in metrics dataframe (#2183 by @lostella)
• Fix recursive case for gluonts.mx.batchify.stack (#2184 by @lostella)
• Fix item_id values in ConstantValuePredictor (#2192 by @codingWhale13)
• Fixup Patience class. (#2197 by @jaheba)
• Fix dataset arrow writer tool. (#2196 by @jaheba)
• Fix SymbolBlock serde issue (#2187 by @lostella)
• Add item id to Uber TLC dataset (#2214 by @mvanness354)
• Fix r_forecast wrapper to shift start date when truncating time series (#2216 by @abdulfatir)
• Fix dtype bug in piecewise_linear and add a test (#2224 by @rshyamsundar)
• Fix bug in to_quantile_forecast (#2225 by @eugeneteoh)
• Fix gluonts.mx.trainer.Trainer in case of empty data loader (#2228 by @lostella)
• Fix feed-forward models when features are provided (#2238 by @lostella)
• update SplicedBinnedPareto demos from nursery version to gluonts version (#2250 by @elenaehrlich)
• Improve len() for ParquetFile. (#2261 by @jaheba)
• Move max_idle_transform usage to GluonEstimator. (#2262 by @jaheba)
• Optimize TimeSeriesSlice performance (#2259 by @lostella)
• Fix ignore hidden files when generating datasets (#2263 by @kashif)
• Fix: set max idle transforms in PyTorch estimators (#2266 by @lostella)
• Fix QuantileForecast.plot() to use DateTimeIndex (#2269 by @abdulfatir)
• Fix serde dataclass eventual. (#2277 by @jaheba)
• Fix gluonts.dataset.split for multivariate case (#2314 by @lostella)
• Improve TestData class in gluonts.dataset.split (#2315 by @lostella)
• Simplify make_evaluation_predictions (#2309 by @lostella)
• Fix MQCNN for kernel_size=1 (#2321 by @lostella)
• Simplify unbatching in forecast-generator. (#2334 by @jaheba)
• Fix numerical bug in BinnedUniforms (#2344 by @moudheus)

Documentation

• Docs: Make notebook templates. (#2122 by @jaheba)
• Docs: Rework installation section. (#2130 by @jaheba)
• Docs: Fix running tutorials for publishing docs. (#2138 by @jaheba)
• Docs: Update hyperparameter tuning with optuna notebook. (#2137 by @npnv)
• Fix issues with hyperparameter tuning tutorial (#2143 by @lostella)
• Apply black to notebooks. (#2144 by @jaheba)
• Docs: Simplify wide DataFrame example (#2150 by @lostella)
• Docs: fix links in models table (#2156 by @lostella)
• Add 'Background' section to docs. (#2129 by @jaheba)
• Docs: Add info about version guarantees. (#2161 by @jaheba)
• Docs: fix tutorial after breaking changes in trainer class (#2179 by @lostella)
• Add tutorial with data splitting examples (#2157 by @npnv)
• Fix: add missing link to splitting tutorial (#2185 by @lostella)
• Fix: ensure last cell of tutorials runs (#2186 by @lostella)
• Fixes to the dataset splitting tutorial (#2189 by @npnv)
• Update TSBench readme with paper reference (#2191 by @geoalgo)
• Update Available models table with the hierarchical model (#2209 by @rshyamsundar)
• Fix broken links in Available-models table (#2211 by @rshyamsundar)
• Add logo to README. (#2248 by @jaheba)
• New logo. (#2243 by @jaheba)
• Use brand colors in docs. (#2257 by @jaheba)
• Docs: Reformatting table, badge colors. (#2258 by @jaheba)
• Docs: update contribution guidelines and dev setup (#2270 by @lostella)
• Add Github footer icon to docs. (#2285 by @jaheba)
• Docs: Custom Pygments style for dark theme. (#2290 by @jaheba)
• Fix README quick examples (#2297 by @lostella)
• Fix text in Quick Start Tutorial (#2300 by @sighellan)
• Update README and tutorial (#2311 by @lostella)
• Turn on apidoc generation (#2332 by @jaheba)
• Add info on how to use 'just' (#2339 by @codingWhale13)
• Small documentation improvements (#2343 by @codingWhale13)

Test / setup changes

• add python 3.9 to test workflows (#2136)
• Tests: Move mx model test. (#2158 by @jaheba)
• Test: Use spawn method for shell server tests. (#2177 by @jaheba)
• Remove holidays and matplotlib from core dependencies. (#2055 by @jaheba)
• Update minimal version for nbconvert. (#2233 by @jaheba)
• Hierarchical: Add a test for to_dataset method (#2265 by @rshyamsundar)
• Fix mypy and black commands in pre-commit githook (#2271 by @abdulfatir)
• Update project_urls. (#2274 by @jaheba)
• Move _version to meta. (#2293 by @jaheba)
• Remove setup-requires. (#2295 by @jaheba)
• Remove pytest.ini. (#2298 by @jaheba)
• Speed up smoke tests (#2341 by @lostella)

Backporting fixes:

• Add Github footer icon to docs. (#2285 by @jaheba)
• Docs: Custom Pygments style for dark theme. (#2290 by @jaheba)
• Fix README quick examples (#2297 by @lostella)
• Fix text in Quick Start Tutorial (#2300 by @sighellan)
• Update README and tutorial (#2311 by @lostella)
• Fix MQCNN for kernel_size=1 (#2321 by @lostella)

Backporting fixes:

• Improve len() for ParquetFile. (#2261 by @jaheba)
• Max idle transform fix (#2262 by @jaheba)
• Fix ignore hidden files when generating datasets (#2263 by @kashif)
• Fix: set max idle transforms in PyTorch estimators (#2266 by @lostella)
• Fix QuantileForecast.plot() to use DateTimeIndex (#2269 by @abdulfatir)

Backporting fixes:

• Fix broken links in Available-models table (#2211 by @rshyamsundar)
• Fix r_forecast wrapper to shift start date when truncating time series (#2216 by @abdulfatir)
• Improve arrow reading performance (#2217 by @mr-1993)
• Fix dtype bug in piecewise_linear and add a test (#2224 by @rshyamsundar)
• Fix bug in to_quantile_forecast (#2225 by @eugeneteoh)
• Fix gluonts.mx.trainer.Trainer in case of empty data loader (#2228 by @lostella)
• Fix feed-forward models when features are provided (#2238 by @lostella)

Full changelog: https://github.com/awslabs/gluon-ts/compare/v0.10.4...v0.10.5

Backporting fixes:

• Fix r_forecast wrapper to shift start date when truncating time series (#2216 by @abdulfatir)
• Fix dtype bug in piecewise_linear and add a test (#2224 by @rshyamsundar)
• Fix bug in to_quantile_forecast (#2225 by @eugeneteoh)
• Fix gluonts.mx.trainer.Trainer in case of empty data loader (#2228 by @lostella)
• Fix feed-forward models when features are provided (#2238 by @lostella)

Full Changelog: https://github.com/awslabs/gluon-ts/compare/v0.9.8...v0.9.9

Backporting fixes:

• Fix SymbolBlock serde issue (#2187 by @lostella)
• Fix dataset arrow writer tool. (#2196 by @jaheba)
• Expose evaluator aggregation_strategy functions (#2198 by @kashif)
• Update Available models table with the hierarchical model (#2209 by @rshyamsundar)

Backporting fixes:

• Fix SymbolBlock serde issue (#2187 by @lostella)

Backporting fixes:

• Fix Prophet wrapper to work with Timestamp instead of Period (#2182 by @lostella)
• Fix dtype for "item_id" column in metrics dataframe (#2183 by @lostella)
• Fix recursive case for gluonts.mx.batchify.stack (#2184 by @lostella)
• Fix: ensure last cell of tutorials runs (#2186 by @lostella)
• Fix item_id values in ConstantValuePredictor (#2192 by @codingWhale13)

Backporting fixes:

• Fix dtype for "item_id" column in metrics dataframe (https://github.com/awslabs/gluon-ts/pull/2183 by @lostella)
• Fix recursive case for gluonts.mx.batchify.stack (https://github.com/awslabs/gluon-ts/pull/2184 by @lostella)
• Fix item_id values in ConstantValuePredictor (https://github.com/awslabs/gluon-ts/pull/2192 by @codingWhale13)

Backport fixes:

• Make PandasDataset faster (#2148 by @lostella)
• Interrupting mx.Trainer stops training. (#2131 by @Hongqing-work)
• Ignore divide warnings in evaluation. (#2159 by @jaheba)

Backporting fixes:

• Docs: Make notebook templates. (#2122 by @jaheba)
• Docs: Rework installation section. (#2130 by @jaheba)
• Fix DatasetCollection for Python 3.9. (#2135 by @rsnirwan)
• Docs: Fix running tutorials for publishing docs. (#2138 by @jaheba)
• Fix PandasDataset for Python 3.9 (#2141 by @lostella)
• Fix issues with hyperparameter tuning tutorial (#2143 by @lostella)
• Docs: Apply black to notebooks. (#2144 by @jaheba)

Overview

Arrow based datasets

We have added support for Parquet-files, as well as Arrow's binary format. This is an opt-in feature, requiring pyarrow to be installed. Use pip install 'gluonts[pro]' or pip install 'gluonts[arrow]' to ensure the correct version is installed.

FileDataset has been reworked to support .parquet and .arrow files. Previously, it had assumed all files to use jsonlines. To continue using jsonlines ensure that the the files use one of the .json, .jsonl, .json.gz, jsonl.gz suffixes.

Depending on the dataset size and shape, Arrow can be much faster than the json variant. In more extreme cases we saw speedups of more than 100x when using arrow vs jsonlines (see #2003 for some examples).

To convert a given dataset into arrow, you can use the gluonts.dataset.arrow utility:

python -m gluonts.dataset.arrow write </path/to/dataset> my-dataset.arrow

PandasDataset

We have added support for pandas.DataFrame and pandas.Series as well. You can now directly model data given in a DataFrame using gluonts.dataset.pandas.PandasDataset. In this tutorial we describe in depth how you can use PandasDataset to speed up modelling using GluonTS.

Changelog

New Features

• #1631 - Add TimeLimitCallback to mx/trainer callbacks. (by @yx1215)
• #1780 - adding MQF2 (Multi-horizon) (by @KelvinKan)
• #1903 - Added QuarterlyBegin time feature (by @kashif)
• #1924 - Porting SimpleFeedForwardEstimator to PyTorch (by @lostella)
• #1925 - DeepAR PyTorch: make samplers configurable (by @lostella)
• #1935 - added support for pandas dataframes (by @rsnirwan)
• #1962 - Add support for beta-NLL loss (by @kashif)
• #1982 - Add Uber-TLC dataset to dataset repository. (by @Hongqing-work)
• #1990 - Add info cli. (by @jaheba)
• #1987 - Add HP tuning example with Optuna (by @npnv)
• #2000 - Add arrow-based dataset. (by @vafl, @lostella, @jaheba)
• #2002 - add ND for item_metrics (by @melopeo)
• #2006 - Added support of "long" RTS, making short RTS be "past_feat_dynamic_real" (by @zoolhasson)
• #2061 - Add DatasetWriter. (by @jaheba)
• #2074 - Add support for second frequency. (by @kashif)

Breaking Changes

• #1917 - Breaking: Fix return types of features (by @lostella)
• #1941 - Breaking: Update dependency fbprophet -> prophet (by @lostella)
• #1946 - Breaking: Split incremental quantile output into separate class (by @lostella)
• #1965 - Breaking: reorg torch package, shorten import paths (by @lostella)
• #1980 - Use pd.Period instead of pd.Timestamp. (by @jaheba)
• #1997 - Remove freq argument from Forecast. (by @kashif)
• #2011 - Remove dct_reduce. (by @jaheba)
• #2017 - Remove mandatory freq attribute of Predictor. (by @kashif)
• #2018 - Remove multiprocessing dataloader. (by @jaheba)
• #2019 - Rework FileDataset. (by @jaheba)
• #2053 - Add dataset_writer to get_dataset. (by @Hongqing-work)
• #2070 - Add jsonl.encode_json, remove serialize_data_entry. (by @jaheba)

Bug Fixes / Minor Improvements

• #1704 - Settings._let will pop element it added instead of just the last one. (by @jaheba)
• #1905 - Fix typing issues in torch estimators, update base estimators docstrings (by @lostella)
• #1909 - Fix the use of the scaling parameter in Transformer model (by @StanislasGuinel)
• #1916 - Fix AddTimeFeatures transformation for multiples of base frequencies (by @lostella)
• #1920 - Fix: use broadcast_lesser in place of comparisons in ISQF (by @vincentqb)
• #1931 - Fix dummy estimator (by @canerturkmen)
• #1933 - Fix Pytorch Lightning tutorial. (by @jaheba)
• #1938 - Fixed autograd inplace operations error in Transformed Distribution (by @shubhamkapoor)
• #1950 - Fix: Hard threshold positive distribution parameters (by @lostella)
• #1952 - Fix forecast keys (quantiles) output by TemporalFusionTransformer (by @lostella)
• #1968 - Fix: use of num_parallel_samples in deepAR (by @kashif)
• #1969 - Fix: torch DeepAR observed indicator in multivariate case (by @kashif)
• #1975 - use FieldName (by @kashif)
• #1983 - Documentation: add docstrings for torch-based models (by @lostella)
• #1986 - Fix OffsetSplitter for negative offsets (by @lostella)
• #1989 - Pin protobuf version. (by @jaheba)
• #1991 - Remove packaged pytorch-ts from gluonts.nursery.SCott (by @lostella)
• #1999 - Documentation: fix and speed up tutorials (by @lostella)
• #2004 - Refactor splitter assertion and add error message (by @RSNirwan)
• #2005 - Rework itertools, add col-to-row and row-to-col functions. (by @jaheba)
• #2008 - Re-add cache for parsing 'pd.Period'. (by @jaheba)
• #2013 - Update website template, clean up homepage and tutorials (by @lostella)
• #2014 - Expose Estimator, Predictor, Forecast in gluonts.model. (by @jaheba)
• #2015 - Fix mean in AffineTransformedDistribution (by @stailx)
• #2016 - Fix torch affine transformed distribution (by @lostella)
• #2020 - Remove unnecessary files from docs folder, update gitignore (by @lostella)
• #2021 - Update references to dev branch. (by @lostella)
• #2024 - Fix README. Use DataFramesDataset. (by @jaheba)
• #2025 - Make HP tuning tutorial more accurate (by @jaheba)
• #2028 - Re-add support for Python 3.6 (by @jaheba)
• #2029 - Add support for nan values in Rotbaum (by @zoolhasson)
• #2035 - Simplify lag values computation in torch DeepAR (by @lostella)
• #2036 - Minor improvements to the hierarchical model (by @rshyamsundar)
• #2047 - Make Quantile derive from pydantic.BaseModel. (by @jaheba)
• #2050 - Add concepts section to docs. (by @jaheba)
• #2051 - Add tutorial on DataFramesDataset (by @RSNirwan)
• #2057 - Add optional parameter time_axis to forecast_start. (by @melopeo)
• #2062 - Fix type annotations for predict_to_numpy (by @lostella)
• #2066 - Always pass freq explicitly to pd.period_range. (by @kashif)
• #2068 - Docs: simplify call to evaluator (by @lostella)
• #2092 - Fix: DistributionLoss not encodable. (by @jaheba)
• #2098 - Add Airtraffic dataset. (by @jaheba)
• #2108 - Fixup trainer in case of non-finite loss. (by @jaheba)
• #2121 - Change default behavior for TrainDatasets overwrite (by @nklingen)

Backporting fixes:

• Fix: DistributionLoss not encodable (#2092 by @jaheba)

Overview

Arrow based datasets

We have added support for Parquet-files, as well as Arrow's binary format. This is an opt-in feature, requiring pyarrow to be installed. Use pip install 'gluonts[pro]' or pip install 'gluonts[arrow]' to ensure the correct version is installed.

FileDataset has been reworked to support .parquet and .arrow files. Previously, it had assumed all files to use jsonlines. To continue using jsonlines ensure that the the files use one of the .json, .jsonl, .json.gz, jsonl.gz suffixes.

Depending on the dataset size and shape, Arrow can be much faster than the json variant. In more extreme cases we saw speedups of more than 100x when using arrow vs jsonlines (see #2003 for some examples).

To convert a given dataset into arrow, you can use the gluonts.dataset.arrow utility:

python -m gluonts.dataset.arrow write </path/to/dataset> my-dataset.arrow

PandasDataset

We have added support for pandas.DataFrame and pandas.Series as well. You can now directly model data given in a DataFrame using gluonts.dataset.pandas.PandasDataset. In this tutorial we describe in depth how you can use PandasDataset to speed up modelling using GluonTS.

Changelog

New Features

• #1631 - Add TimeLimitCallback to mx/trainer callbacks. (by @yx1215)
• #1780 - adding MQF2 (Multi-horizon) (by @KelvinKan)
• #1903 - Added QuarterlyBegin time feature (by @kashif)
• #1924 - Porting SimpleFeedForwardEstimator to PyTorch (by @lostella)
• #1925 - DeepAR PyTorch: make samplers configurable (by @lostella)
• #1935 - added support for pandas dataframes (by @rsnirwan)
• #1962 - Add support for beta-NLL loss (by @kashif)
• #1982 - Add Uber-TLC dataset to dataset repository. (by @Hongqing-work)
• #1990 - Add info cli. (by @jaheba)
• #1987 - Add HP tuning example with Optuna (by @npnv)
• #2000 - Add arrow-based dataset. (by @vafl, @lostella, @jaheba)
• #2002 - add ND for item_metrics (by @melopeo)
• #2006 - Added support of "long" RTS, making short RTS be "past_feat_dynamic_real" (by @zoolhasson)
• #2061 - Add DatasetWriter. (by @jaheba)
• #2074 - Add support for second frequency. (by @kashif)

Breaking Changes

• #1917 - Breaking: Fix return types of features (by @lostella)
• #1941 - Breaking: Update dependency fbprophet -> prophet (by @lostella)
• #1946 - Breaking: Split incremental quantile output into separate class (by @lostella)
• #1965 - Breaking: reorg torch package, shorten import paths (by @lostella)
• #1980 - Use pd.Period instead of pd.Timestamp. (by @jaheba)
• #1997 - Remove freq argument from Forecast. (by @kashif)
• #2011 - Remove dct_reduce. (by @jaheba)
• #2018 - Remove multiprocessing dataloader. (by @jaheba)
• #2019 - Rework FileDataset. (by @jaheba)
• #2053 - Add dataset_writer to get_dataset. (by @Hongqing-work)
• #2070 - Add jsonl.encode_json, remove serialize_data_entry. (by @jaheba)

Bug Fixes / Minor Improvements

• #1704 - Settings._let will pop element it added instead of just the last one. (by @jaheba)
• #1905 - Fix typing issues in torch estimators, update base estimators docstrings (by @lostella)
• #1909 - Fix the use of the scaling parameter in Transformer model (by @StanislasGuinel)
• #1916 - Fix AddTimeFeatures transformation for multiples of base frequencies (by @lostella)
• #1920 - Fix: use broadcast_lesser in place of comparisons in ISQF (by @vincentqb)
• #1931 - Fix dummy estimator (by @canerturkmen)
• #1933 - Fix Pytorch Lightning tutorial. (by @jaheba)
• #1938 - Fixed autograd inplace operations error in Transformed Distribution (by @shubhamkapoor)
• #1950 - Fix: Hard threshold positive distribution parameters (by @lostella)
• #1952 - Fix forecast keys (quantiles) output by TemporalFusionTransformer (by @lostella)
• #1968 - Fix: use of num_parallel_samples in deepAR (by @kashif)
• #1969 - Fix: torch DeepAR observed indicator in multivariate case (by @kashif)
• #1975 - use FieldName (by @kashif)
• #1983 - Documentation: add docstrings for torch-based models (by @lostella)
• #1986 - Fix OffsetSplitter for negative offsets (by @lostella)
• #1989 - Pin protobuf version. (by @jaheba)
• #1991 - Remove packaged pytorch-ts from gluonts.nursery.SCott (by @lostella)
• #1999 - Documentation: fix and speed up tutorials (by @lostella)
• #2004 - Refactor splitter assertion and add error message (by @RSNirwan)
• #2005 - Rework itertools, add col-to-row and row-to-col functions. (by @jaheba)
• #2008 - Re-add cache for parsing 'pd.Period'. (by @jaheba)
• #2013 - Update website template, clean up homepage and tutorials (by @lostella)
• #2014 - Expose Estimator, Predictor, Forecast in gluonts.model. (by @jaheba)
• #2015 - Fix mean in AffineTransformedDistribution (by @stailx)
• #2016 - Fix torch affine transformed distribution (by @lostella)
• #2020 - Remove unnecessary files from docs folder, update gitignore (by @lostella)
• #2021 - Update references to dev branch. (by @lostella)
• #2024 - Fix README. Use DataFramesDataset. (by @jaheba)
• #2025 - Make HP tuning tutorial more accurate (by @jaheba)
• #2028 - Re-add support for Python 3.6 (by @jaheba)
• #2029 - Add support for nan values in Rotbaum (by @zoolhasson)
• #2035 - Simplify lag values computation in torch DeepAR (by @lostella)
• #2036 - Minor improvements to the hierarchical model (by @rshyamsundar)
• #2047 - Make Quantile derive from pydantic.BaseModel. (by @jaheba)
• #2050 - Add concepts section to docs. (by @jaheba)
• #2051 - Add tutorial on DataFramesDataset (by @RSNirwan)
• #2057 - Add optional parameter time_axis to forecast_start. (by @melopeo)
• #2062 - Fix type annotations for predict_to_numpy (by @lostella)
• #2068 - Docs: simplify call to evaluator (by @lostella)

• Re-add support for Python 3.6 in v0.9.x. (#2032 by @jaheba)

Backporting fixes:

• Fix: use of num_parallel_samples in deepAR (#1968 by @kashif)
• Fix: torch DeepAR observed indicator in multivariate case (#1969 by @kashif)
• Fix OffsetSplitter for negative offsets (#1986 by @lostella)
• Fix mean in AffineTransformedDistribution (#2015 by @stailx)