Nik

> I'm not sure I fully understand the questions. > > > Why does predict not actually predict the next n timesteps? > > It does. If you do `print(len(naive_forecast))`...

Hi, I never got around to trying the nixtla package because I've been having good success with another top forecasting method/package (TFT from the [pytorch forecasting package](https://pytorch-forecasting.readthedocs.io/en/stable/index.html)) However, I have...

Annoying error/bug that happened for me when I trained a TFT model on a GPU workstation with recent PyTorch & PyTorch Forecasting (1.12.1 & 0.10.2) and tried to deploy the...

I was faced with this problem of needing to adjust my vertical space when using `guides(colour=guide_legend(ncol=2,byrow=FALSE))`. As you have described it is not possible to adjust the vertical spacing between...

Thank you for your fast reply. My problem with the `GapRollForward` is that - without adjusting/tuning the various input arguments - it lead to overfitting when finetuning my time series...

Thank you for your comments. I like your code and see that it is very similar to the Rep-Holdout; just not randomized and instead truly rolling forward.

Thank you for the insights! What are your thoughts on the paper I linked to then and the performance of the different methods?

I also just fill the empty cells with an invisible character (`\,` works for Latex). A fix for this bug would be appreciated however.

Did you solve this problem? I also need this change and will work on it on my fork if no one has implemented it already.

I added this functionality to my [fork](https://github.com/georgeblck/ggradar) You should be able to install this version with ```R devtools::install_github("georgeblck/ggradar") ```