Add regressors to the model

[Kate209]

Hi, Matthijs!

You’ve done a good job with your timeseers package! Thanks! I’m a data scientist in a big retail company and we are now working on a new product in which we want to forecast sales. It seems like timeseers could be a great decision because we have lots of SKUs and it is really difficult to builld a model for each of them individually. In our model we would like to use some exogenous regressors. It seems like timeseers already has a “Regressor” module. However, it is not very clear if it works and how it works. Could you please give some comments about it or some toy example how to use it? Does it work only with categorical features or continuous features can also be used? What “ON” argument is used for?

Thanks in advance for your reply!

Kate

[MBrouns]

Hi Kate!

Thanks for showing interest, and glad you like my ideas. Exogenous regressors are indeed sort of part of the library. I think the actual modelling part of it probably works, but the prediction API I built initially doesn't support them very well. I made some steps into a new prediction API that would work nicely with exogenous variables but I got sidetracked a bit. I have some time this weekend so I might just finish it up and work up an example of the regressor component.

[Kate209]

Thanks for quck reply! Looking forwad to updates :)

[Kate209]

Hi Matthijs!

I've tried to make my own version of regression component for partial pooling. Could you please look though my example code if you have some time? And maybe you can give some comments about whether I am doing it correctly?

I craeted a toy dataset with two timeseries (ID=0 and ID=1), each of length 200. It was constructed as seasonality1 + seasonality2 + trend + regresion component + some noise. Then I fitted the model with 3 components: trend, seasonality and regressor. The code and results are in the jupyter notebook. demo.zip

Thanks in advance!

[MBrouns]

I just took a quick look and I think what you're doing makes sense. As it stands though your regressors parameters are completely unpooled, is that what you wanted?

Also you're getting a lot of divergences. I've had the same happen when the data I was fitting was too "perfect" for the model. In those cases the likelihood of the exact right parameters is so much higher than any other value that the slope of that likelihood also becomes very high, causing divergent transitions. Fitting on a less perfect dataset can actually help speed up the fitting process in these cases