Autoregressive models on time series data can be a challenging when it comes to complex distribution shift dynamics and noise.
This repo features a synthetic AR (autoregressive) generator, where we can control many parameters of the AR model. It basically allows us to simulate an AR process. By using this, one can introduce noise and/or control other parameters, of a time-series dataset.
Using a synthetic dataset helps evaluate different machine learned models for noise and dynamic distribution shifts in time series data.
The notebook consists of two parts.
- Data Generation without regression
- Uses the module "timeseries-generator", to help user visualize parameter tweaks. The last cell should be the one that actually generates the synthetic data file.
- Data Generation with regression.
- Uses ARMA-simulator from https://github.com/igarizio/ARMA-simulator/blob/HEAD/ARMA.py.
https://otexts.com/fpp2/arima.html
https://www.seanabu.com/2016/03/22/time-series-seasonal-ARIMA-model-in-python/
https://towardsdatascience.com/creating-synthetic-time-series-data-67223ff08e34
https://pypi.org/project/timeseries-generator/
https://github.com/igarizio/ARMA-simulator/blob/HEAD/ARMA.py