etna 2.9.0

Highlights

• Add possibility to load pretrained embedding models
• Add is_freezed parameter to TS2VecEmbeddingModel and TSTCCEmbeddingModel
• Convert segment to string during reading csv in backtest and forecast commands
• Fix holidays during loading datasets traffic_2008_10T and traffic_2008_hourly
• Fix ModelDecomposeTransform import without prophet module

Full changelog

Added

• Add **kwargs argument description for models based on LinearRegression, ElasticNet and CatBoostRegressor (#454)
• Add possibility to load pretrained embedding models (#461)
• Add is_freezed parameter to TS2VecEmbeddingModel and TSTCCEmbeddingModel (#461)
• Add test on working without extras (#463)

Changed

• Add support of property attributes in __repr__ and to_dict of BaseMixin (#469)

Fixed

• Fix IForestOutlierTransform failed with ignored target column (#460)
• Add lower limit for typing_extension versions (#458)
• Fix ModelDecomposeTransform import without prophet module (#459)
• Convert segment to string during reading csv in backtest and forecast commands (#470)
• Fix holidays during loading datasets traffic_2008_10T and traffic_2008_hourly (#462)

etna 2.8.0

Highlights

• Add new anomaly detection functions: get_anomalies_mad, get_anomalies_iqr, get_anomalies_isolation_forest
• Add new anomaly detection transforms: MADOutlierTransform, IForestOutlierTransform, IQROutlierTransform
• Add decomposition transforms for anomaly detection: FourierDecomposeTransform ModelDecomposeTransform
• Add MeanEncoderTransform
• Set upper bound <2 on numpy version
• Fix VotingEnsemble, StackingEnsemble, DirectEnsemble to have a valid params_to_tune that returns empty dict

Full changelog

Added

• Add get_anomalies_iqr function for anomaly detection (#374)
• Add get_anomalies_isolation_forest method for anomaly detection (#375)
• Add IForestOutlierTransform (#381)
• Add IQROutlierTransform (#387)
• Add num_workers parameter to TS2VecEmbeddingModel (#396)
• Add get_anomalies_mad function for anomaly detection (#398)
• Add TSDataset.features property to get list of all features in a dataset (#405)
• Add MADOutlierTransform class for anomaly detection (#415)
• Add MeanEncoderTransform (#413)
• Add FourierDecomposeTransform transform for series decomposition using DFT (#430)
• Add ModelDecomposeTransform transform for series decomposition using ETNA models (#427)

Changed

• Allow to change device, batch_size and num_workers of embedding models (#396)
• Update pipelines documentation (#408)
• Update formulas for metrics in documentation (#406)
• Update documentation to explain how to contribute and work with discussions, update templates for issues (#395)
• Remove "Other issue" template, update links to discussions in issue creation menu (#401)

Fixed

• Fix rendering in 210 tutorial (#386)
• Fix typo in 103 tutorial (#408)
• Remove sorting of ts.df by timestamps in plot_forecast and plot_forecast_decomposition (#410)
• Fix forecast visualization with horizon=1 (#426)
• Set upper bound <2 on numpy version (#431)
• Fix VotingEnsemble, StackingEnsemble, DirectEnsemble have a valid params_to_tune that returns empty dict (#432)
• Fix passing custom model to STLTransform (#412)
• Update TSDataset.describe, TSDataset.info to exclude target intervals and target components in num_exogs (#405)

etna 2.7.1

Highlights

• Fix errors when importing modules without torch extras (#382)

Full changelog

Fixed

• Fix errors when importing modules without torch extras (#382)

etna 2.7.0

Highlights

• Add EmbeddingSegmentTransform
• Add EmbeddingWindowTransform
• Add tutorial 210-embedding_models about using embedding models
• Add parameter drop_zero into MRMRFeatureSelectionTransform
• Allow RNNModel, MLPModel, DeepARNativeModel, DeepStateModel to work with categorical features
• Allow encoders to return numeric features

Full changelog

• Add TS2VecEmbeddingModel model (#253)
• Add EmbeddingSegmentTransform (#265)
• Add EmbeddingWindowTransform (#265)
• Add TSTCCEmbeddingModel (#294)
• Add 210-embedding_models example notebook (#304)
• Add parameter drop_zero into MRMRFeatureSelectionTransform (#308)

Changed

• Allow RNNModel to work with categorical features (#334)
• Allow DeepARNativeModel and MLPModel to work with categorical features (#336)
• Allow DeepState to work with categorical features (#342)
• Allow encoders to return numeric features (#352)
• Enable cancelling old CI/CD runs after changes in a branch (#339)

Fixed

• Fix FordA download url in classification notebook (#309)
• Allow seaborn dependency to have higher version (#319)
• Fix MRMRFeatureSelectionTransform to correctly handle less-is-better relevance_table (#308)
• Fix PatchTSModel fails when using additional features (#376)
• Fix 101-get-started notebook to be rendered correctly (#340)
• Fix DeepStateModel forecasting problem with horizon=1 (#377)

etna 2.6.0

Highlights

• Add ability to work with integer timestamp by using freq=None
• Add tutorial about working with misaligned data by using integer timestamp
• Add TSDataset.create_from_misaligned constructor
• Add infer_alignment, apply_alignment, make_timestamp_df utils into etna.dataset
• Add BinaryOperationTransform
• Add TFTNativeModel
• Add in OutliersTransform possibilities use ignore_flag_column to skip values use ignore
• Extend base TSDataset constructor to handle long format dataframes
• Speed up timestamp transforms
• Update timestamp transforms to work with external timestamp by using in_column parameter

Full changelog

Added

• Add BinaryOperationTransform to transforms (#260)
• Add TFTNativeModel (#290)
• Add warning on trying to pass numeric timestamp if freq is not None and add _cast_index_to_datetime (#214)
• Add infer_alignment, apply_alignment, make_timestamp_df into etna.dataset.utils (#256)
• Add TSDataset.create_from_misaligned constructor (#269)
• Add tutorial about working with misaligned data (#288)
• Add in OutliersTransform possibilities use ignore_flag_column to skip values use ignore (#291)

Changed

• Update glossary with terms related to working with misaligned data (#288)
• Add ignoring of integer timestamp as a feature into native DL models (#210)
• Update pytorch_forecasting models to handle integer timestamp (#208)
• Update datasets module to work with integer timestamp (#146)
• Add tests for transform on data with integer timestamp (#153)
• Add tests for models on data with integer timestamp (#188)
• Update DateFlagsTransform, TimeFlagsTransform, HolidayTransform, SpecialDaysTransform, FourierTransform to work with external timestamp (#169)
• Update analysis module to work with integer timestamp (#161)
• Update StatsForecastARIMAModel, StatsForecastAutoARIMAModel, StatsForecastAutoCESModel, StatsForecastAutoETSModel, StatsForecastAutoThetaModel to handle integer timestamp (#197)
• Update MRMRFeatureSelectionTransform to handle integer timestamp (#164)
• Update deseasonality transforms (STLTransform, DeseasonalityTransform) to handle integer timestamp (#174)
• Update HoltModel, HoltWintersModel, SimpleExpSmoothingModel, SARIMAXModel, AutoARIMAModel to handle integer timestamp ((#200)[https://github.com//pull/200])
• Update detrend transforms (LinearTrendTransform, TheilSenTrendTransform) to handle integer timestamp (#163)
• Update ResampleWithDistributionTransform to work with integer timestamp (#165)
• Update change point transforms (ChangePointsSegmentationTransform, ChangePointsTrendTransform, ChangePointsLevelTransform, TrendTransform) to handle integer timestamp (#176)
• Update BATSModel, TBATSModel models to work with integer timestamp (#195)
• Update ProphetModel to handle external timestamp (#203)
• Remove checking frequency in timestamp_column of ProphetModel (#222)
• Update FourierTransform to handle external datetime timestamp (#223)
• Update FoldMask to work with integer timestamp, in validate_on_dataset method add validation on presence of FoldMask parameters in ts.index, add tests for FoldMask (#226)
• Fix FourierTransform on integer index, add inference tests (#230)
• Update outliers transforms to handle integer timestamp (#229)
• Update pipelines to handle integer timestamp (#241)
• Add timestamp_range and refactor code with it (#244)
• Update CLI to handle integer timestamp (#246)
• Update ExogShiftTransform to handle integer timestamp (#254)
• Extend base TSDataset constructor to handle long format dataframes, update documentation and tutorials with this change (#266)
• Update internal datasets to work with unaligned data (#292)
• Speed up "timestamp" transforms (#295

Fixed

• Fix PredictionIntervalOutliersTransform fails to work with created columns (#291)
• Prohibit empty list value and duplication of target_timestamps parameter in FoldMask (#226)
• Fix DeseasonalityTransform fails to inverse transform short series (#174)
• Fix indexing in stl_plot, plot_periodogram, plot_holidays, plot_backtest, plot_backtest_interactive, ResampleWithDistributionTransform (#244)
• Fix DifferencingTransform to handle integer timestamp on test (#244)
• Fix HolidayTransform to handle integer timestamp in days_count mode (#285)

etna 2.5.0

Highlights

• Add built-in datasets: M3, M4, electricity, etc. The full list can be viewed at the documentation.
• Add new mode="days_count" to HolidayTransform
• Add method size to TSDataset class
• Optimize performance of DensityOutliersTransform
• Fix method to_dict for SklearnPerSegmentModel and SklearnMultiSegmentModel
• Add the index_only parameter to outlier analysis functions

Full changelog

Added

• Add electricity to internal datasets (#60)
• Add parts argument to load_dataset function (#79)
• Add M4 to internal datasets (#83)
• Add M3 to internal datasets (#91)
• Add traffic_2008 to internal datasets (#94)
• Add traffic_2015 to internal datasets (#100)
• Add tourism to internal datasets (#120)
• Add weather to internal datasets (#125)
• Add ETT to internal datasets (#134)
• Add list_datasets function (#145)
• Add IHEPC to internal datasets (#150)
• Add dataset integrity check using hash for internal datasets (#151)
• Create page about internal datasets in documentation (#175)
• Add usage example of internal datasets in 101-get_started.ipynb and 305-classification.ipynb tutorials (#202)
• Add new mode="days_count" to HolidayTransform (#239)
• Add size method to TSDataset class (#238)
• Add the index_only parameter to outlier analysis functions for return type control (#231)

Changed

• Add relevance_aggregation_mode and redundancy_aggregation_mode into MRMRFeatureSelectionTransform.params_to_tune (#212)
• Optimized DensityOutliersTransform and removed _save_original_values from outlier transforms (#231)
• Update python to 3.10 in CI (#251)

Fixed

• Fix traffic_2008 (128)
• Fix number of segments in docs, column name for tourism dataset and change default save path (#206)
• Fix method to_dict for SklearnPerSegmentModel and SklearnMultiSegmentModel (#199)
• Fix method fit for MRMRFeatureSelectionTransform with redundancy_aggregation_mode=median (#212)
• Fix method predict_components for _CatBoostAdapter working incorrectly on shuffled columns (#227)

etna 2.4.0

Highlights

• Add module etna.experimental.prediction_intervals with new model-agnostic methods to generate prediction intervals
• Add notebook about prediction intervals
• Add DeepARNativeModel
• Optimize TSDataset.to_hierarchical_dataset, BasePipeline._validate_backtest_dataset, datasets.utils.duplicate_data

Full changelog

Added

• Add params_to_tune for DeepStateModel (#115)
• Handle new functionality for prediction intervals in the plot_forecast (#130)
• Add get_historical_forecasts to pipelines for forecast estimation at each fold on the historical dataset (#143)
• ConformalPredictionIntervals method for prediction intervals estimation (#152)
• Add DeepARNativeModel (#114)
• EmpiricalPredictionIntervals method for prediction intervals estimation (#173)
• Prediction intervals tutorial notebook (#189)

Changed

• Change warning condition on loading object saved under different library version (#31)

Fixed

• Speed up segment column creation in TSDataset.to_hierarchical_dataset (#194)
• Speed up BasePipeline._validate_backtest_dataset (#194)
• Speed up datasets.utils.duplicate_data (#194)

etna 2.3.0

Highlights

• Add LimitTransform to forecast within limits
• Add EventTransform to create features before and after the event
• Add parameter save_ts to pipeline method fit
• Rework building docker images to use poetry.lock to make them more stable
• Add quickstart notebook
• Rework get_started notebook
• Add mechanics_of_forecasting notebook
• Add code to generate new documentation

Full changelog

Added

• Handle prediction intervals similar to target components in TSDataset (#97)
• SavePredictionIntervalsMixin for the BasePredictionIntervals (#87)
• Base class BasePredictionIntervals for prediction intervals into experimental module (#86)
• Add fit_params parameter to etna.models.sarimax.SARIMAXModel (#69)
• Add quickstart notebook, add mechanics_of_forecasting notebook (#1343)
• Add gallery of tutorials divided by level (#46)
• Create documentation page with links to external resources (#44)
• Add documentation page with glossary of terms (#45)
• Add publishing into s3 for the latest documentation version (#50)
• Add publishing into s3 during release (#53)
• Add multiversion switcher into documentation (#55)
• Add error page into documentation (#57)
• Add LimitTransform (#63)
• Add config for Codecov to control CI (#80)
• Add EventTransform (#78)
• NaiveVariancePredictionIntervals method for prediction quantiles estimation (#109)
• Update interval metrics to work with arbitrary interval bounds (#113)

Changed

• Refactored transform inversion logic in Pipeline forecast method (#72)
• Add parameter save_ts to pipeline method fit (#73)
• Add installation page and notes about extensions into documentation of public classes (#1339)
• Merge User Guide and API sections in documentation, limit classes to show in API section (#1324)
• Unify example notebooks, rerun example notebooks (#1330)
• Rework get_started notebook (#1343)
• Add missing classes from decomposition into API Reference, add modules into page titles in API Reference (#61)
• Update CONTRIBUTING.md with scenarios of documentation updates and release instruction (#77)
• Set up sharding for running tests (#99)
• Rework saving DL models by separating saving model's hyperparameters and model's weights (#98)
• Deprecated FutureMixin (#58)

Fixed

• Fix ResampleWithDistributionTransform working with categorical columns (#82)
• TSDataset._hierarchical_structure_from_level_columns to support pandas>=1.4,<1.5(#107)
• Fix links from tinkoff-ai/etna to etna-team/etna (#47)
• Fix CI job cron-delete-untagged-images (#95)
• Rendering table of contents in notebooks (#1343)
• Fix formatting of docstrings, fix links from netlify to docs.etna.ai (#62)
• Fix multiple warnings, revert catching warnings during testing (#105)
• Fix bug with numpy.warnings in numpy>=1.24, rework building docker images to use poetry.lock (#116)
• Fix name of steps in publish CI (#119)