Description

In this notebook, you'll find the implementation of exponential smoothing I proposed in my book “Data Science for Supply Chain Forecasting”.

The main objective of these functions was to teach exponential smoothing in my book, not to be efficient. 
Similarly, I share these for you to be able to experiment and learn with pandas, numpy, and exponential smoothing models.
Don't expect a E2E working model.

Here are some ideas to speed up the model,
- Change the implementation from Pandas to Numpy
- Change the 1D implementation to a 2D implementation (that is, run all time series in parallel in a single numpy array)
- You can parallelize the search over the parameters using joblib
- You can get some good speed up using smarter techniques to assess seasonality.

To make the model accurate, I would advise, 
- Refining the parameter search space
- Changing the objective KPIs
- Finding better ideas to assess seasonality
- Maybe even trying out some ensembling ideas

Happy learning and coding!
(And don't hesitate to share your own implementations and ideas)

PS: I shared the code in a Jupyter Notebook, but in practice I don't use Jupyter and I always prefer using .py files and Spyder.