Hello everyone, nice to be here! Big thanks to the...
# neural-forecast
Hello everyone, nice to be here! Big thanks to the team for their contribution(s). Quick question, to be familiar with the package I am planning to use it to model a multi-step multivariate ts problem; any recommendations on which model to try out first? I apologise if this is answered somewhere already.
đŸ‘€ 1
Hello Pandula, thanks for joining! All the models in our neuralforecast library are multi-step. The hyperparameter n_time_out controls the forecast horizon. I recommend trying the N-HiTS or RNN models first. They are univariate, but can be used for multivariate problems as well since they will share parameters for all the ts (one model will be used to forecast all ts). Check our colab example: https://colab.research.google.com/drive/1WjBbQzaivQhOldGolzymOtLmo6QX4Ieg#scrollTo=HXKT2-fpUD0Z. We use the N-HiTS model in a multivariate multi-step problem.
NeuralForecast's models take as input your target data with all your series
and exogenous data
with columns
unique_id, ds
hey thank you both for the help and suggestion. I tried it and got it to work, so that’s a win! I am trying to figure out how to work with
now, specially how to structure it. I guess
is used to differentiate between different series? If I keep the
columns plus all other exogenous vars in the
, it would work right? (apologies for not trying it out and asking, it’s pretty late and I only saw this message now) I scanned through the notebooks in
and saw examples, would they be the best source of documentation to refer to?
@Cristian (Nixtla) Also, when you say
one model will be used to forecast all ts)
does it mean the model act similar to VAR models or each ts is separately forecasted as univariate ts?
each ts is separately forecasted
Thanks @Cristian (Nixtla)!