Olivier Sprangers

@ncooder Thanks for the suggestion, you are very right for suggesting this - it's my top priority of next feature to add.

> Hi, will THIEF and MAPA be available soon in Nixtla? Those are really breaking methods sometimes. Thanks for all your work. Target is next two months.

Hey @carusyte, sorry for the late response. Thank you for your contribution. Could you please [follow the contributing guide](https://github.com/Nixtla/neuralforecast/blob/main/CONTRIBUTING.md) when submitting a PR?

Do you also experience this issue with other models? (e.g. `TSMixer`? or `DLinear`?)

@lastsummerx Thanks for that - makes sense. We already have a PR that implements that, we might do it there

Hi, apologies for the late reply. [This tutorial](https://docs.nixtla.io/docs/tutorials-exogenous_variables) can be followed - it specifically addresses the issue where one doesn't have the future values of the exogenous. Hope this helps.

@yeongnamtan #453 will add the capability for including historical exogenous variables. The tutorials will be updated accordingly too.

Hi, apologies for the late reply. You can achieve what you want by specifying a large number of quantiles, for example: `loss = DistributionLoss(distribution='StudentT', quantiles=np.linspace(0, 1, 100)) ` This will...

As of 3.0.0, you can specify a quantiles argument in predict, i.e. nf.predict(...., quantiles=[0.1, 0.2,....]) this works for DistributionLosses and point losses with conformal intervals. So it's similar but slightly...

Sure, we could add a sample endpoint, don't think that's too hard. I'll reopen the issue. Although not sure if you sample and do it e.g. 200 times to compute...