Change Log

[1.2.2] - 2022-02-09

Bump dependencies
Managed different import mechanisms for Dash

[1.2.1] - 2021-06-23

Removed automatic Seasonal part of ARIMA; it could take a long time. Further investigation on pmdarima is requested;
Fixed a bug in which CSV with dates in the format 1959-01 were not correctly recognized;
Fixed a bug in Yeo-Johnson transformation which would "lose" the datetime index of transformed data;
Bump dependencies.

[1.2.0] - 2021-04-08

Added

Exponential Smoothing model, with automatic selection of seasonality and inner model
Options to set max and min values for predictions (if upper/lowerbounds are known beforehand)
Option to round the predictions to the nearest integer
Added _all key to set additional regressors for all the time-series in a dataset

Changed

Multithreading now uses Joblib: it should work on all platforms, and also for NeuralProphet/LSTM models
Fix bug: cross-correlation were wrong if max_lags was higher than the length of the dataset
Models can be specified in the param_config dictionary using lowercase names
Bump dependencies

[1.1.0] - 2021-03-08

Added

Resizable histograms in data visualization
Historical error plot and histogram with percentual
Buttons to quickly change box and aggregate box plots period
Added standard deviation of error in error metrics
New options for model_parameters: min_values/max_values which can be used to specify minimum and maximum values to use in prediction
New option for model_parameters: round_to_integer useful to point columns'names which should be predicted as integer numbers

Changed

When using max_threads: 1 multiprocessing is completely ignored, this temporarily fix some issues on MacOs. Multiprocessing should be rewritten to work on all platforms.

[1.0.0] - 2021-02-17

Initial release.