0.3.6

Latest

• back_forecast for forecast on training data
• Mosaic ensembles can now be used beyond training forecast_length and for shorter lengths too
• best_model_name, best_model_params, and best_model_transformation_params AutoTS attributes now available
• mean, median, and ffill NaN now handle fully NaN series by returning 0.
• fixed bug that was causing mosaic generalization to fail if ffill/bfill handled all missing values
• STLFilter and HPFilter and convolution_filter Transformers added

0.3.5

Latest

• New Transfromer ScipyFilter
• New models Univariate and MultivariateMotif
• 'midhinge' and "weighted_mean" to AverageValueNaive
• Add passing regressors to WindowRegression and made more efficient window generation
• more plotting methods: plot_horizontal_transformers
• for most -Regression type models, model_params is now treated as kwargs and can accept any args for that model
• ExtraTrees and RadiusRegressor to -Regression type models
• bug fix in generate_score_per_series
• 'Generation' now tracked in results table, plus plotting method for generation loss

0.3.4

Latest

• improvements to joblib parallelized models (not copying the full df)
• additonal parameter checks
• made "auto" cpu_count even more conservative
• improved 'Score' generation. It should now be more equally weighted across metrics.
• fixed potential bug for horizontal ensemble selection if perfect forecasts were delivered
• Horizontal ensembles now chosen by combination of multiple metrics and metric_weighting (mae, rmse, spl, contour)
• re-weighted fillna probabilities for random choice
• addressed a few deprecation warnings
• new plot_horizontal() function for AutoTS to quickly visual horizontal ensembles
• Probabilistic and HDist ensembles are now deprecated (they can still be run by model_forecast but not by AutoTS class)
• new introduce_na parameter which makes series more robust to the last values being NaN in final but never in any validation
• Mosaic Ensembles! These can offer major improvements to MAE, but are also less stable than horizontal ensembles.

0.3.3

Latest

• Fixed horizontal ensembles running in univariate cases (they are explicitly multivariate)
• 'superfast' transformer list added
• test on Mac for the first time, everything seems to work except lightgbm
• include first actual unittests (from existing test.py runs)
• slight change to random template generation to make sure all models are choosen at least once
• cleaned up PredictWitch -> model_forecast() a bit so that users can use it to run single models from parameters directly
• added load_live_daily() example data and spruced up production_example.py
• tried in vain to make a quiet verbosity option for GluonTS
• added create_lagged_regressor
• added Greykite model (additional regressors not working yet)
• fixed regressors bug in Prophet
• added a simple plot method to PredictionObject
• fix for deprecation warning in GLS

0.3.2

Latest

• Table of Contents to Extended Tutorial/Readme.md
• Production Example
• add weights="mean"/median/min/max
• UnivariateRegression
• fix check_pickle error for ETS
• fix error in Prophet with latest version
• VisibleDeprecation warning for hidden_layers random choice in sklearn fixed
• prefill_na option added to allow quick filling of NaNs if desired (with zeroes for say, sales forecasting)
• made horizontal generalization more stable
• fixed bug in VAR where failing on data with negatives

0.3.1

Latest

• Additional models to GluonTS
• GeneralTransformer transformation_params - now handle None or empty dict
• cleaning up of the appropriately named 'ModelMonster'
• improving MotifSimulation
• better error message for all models
• enable histgradientboost regressor, left it out before thinking it wouldn't stay experimental this long
• import_template now has slightly better method input style
• allow ensemble parameter to be a list
• NumericTransformer
  • add .fit_transform method
  • generally more options and speed improvement
• added NumericTransformer to future_regressors, should now coerce if they have different dtypes

0.3.0

Latest

• breaking change to model templates: transformers structure change
  • grouping no longer used
• parameter generation for transformers allowing more possible combinations
• transformer_max_depth parameter
• Horizontal Ensembles are now much faster by only running models on the subset of series they apply to
• general starting template improved and updated to new transformer format
• change many np.random to random
  • random.choices further necessitates python 3.6 or greater
• bug fix in Detrend transformer
• bug fix in SeasonalDifference transformer
• SPL bug fix when NaN in test set
• inverse_transform now fills NaN with zero for upper/lower forecasts
• expanded model_list aliases, with dedicated module
• bug fix (creating 0,0 order) and tuning of VARMAX
• Fix export_template bug
• restructuring of some lower-level function locations

0.2.8

Latest

• Round transformer to replace coerce_integer, ClipOutliers expanded, Slice to replace context_slicer
• pd.df Interpolate methods added to FillNA options, " " to "_" in names, rolling_mean_24
• slight improvement to printed progress messages
• transformer_list (also takes a dict of value:probability) allows adjusting which transformers are created in new generations.
  • this does not apply to transformers loaded from imported templates

0.2.7

Latest

• 2x speedup in transformation runtime by removing double transformation
• joblib parallel to UnobservedComponents
• ClipOutliers transformer, Discretize Transformer, CenterLastValue - added in prep for transform template change
• bug fix on IntermittentOccurence
• minor changes to ETS, now replaces single series failure with zero fill, damped now is damped_trend
• 0.3.0 is expected to feature a breaking change to model templates in the transformation/pre-processing

0.2.6

Latest

• fix verbose > 2 error in auto_model
• use of f-strings to print some error messages. Python 3.5 may see more complicated error messages as a result.
• improved BestN (formery Best3) Ensembles, ensemble collected in dicts
• made Horizontal and BestN ensembles tolerant of a component model failure
• made Horizontal models capable of generalizing from a subset of series
• added info to model table for models that can use future_regressor
• added Datepart Regression model, sklearn regressor on time components only