eschibli

I'd be surprised if this explained such a large discrepancy, but while it wasn't clear in their paper the TiDE authors used [`temporal_hidden_size_past=64`](https://github.com/unit8co/darts/issues/2408).

Should a warning be raised if `activation` is None and `hidden_fc_sizes` isn't? That would always be suboptimal and users who don't read the changelog might not be aware the default...

If `multi_models=True` (the default), `output_chunk_length` different, independent linear models are trained, one for each horizon up to `output_chunk_length`.

Madtoinou, why are the quantiles so wide then? Wouldn't 1000 identical predictions imply zero uncertainty?

As implemented in DARTS the TiDE model will use `input_chunk_length` for both `lags` and `lags_past_covariates` as defined in the regression models and will correspondingly use `output_chunk_length` for `lags_future_covariates`. Nearly all...

I'm not a developer of the library myself, but it is near-ubiquitous for neural network-based forecasting models to define a single lookback for past target values and covariates, and a...

Why don't we add an `index_from` argument? This could default to `None` for current behavior, but accept an integer or string index for the component to which the index should...

Alright, I understand the suggestion to use `DataFrames` for analysis (and I admit I initially didn't consider the case of multivariate `TimeSerieses`.) But on further reflection I still think the...

No, it cannot. RIN is effectively a sample-wise scaler - it sets the lookback period of each input to mean 0, variance 1 for each train and inference sample, then...

You should be able to achieve what you want by using `output_chunk_shift`, which sets the gap between the last observation and the first prediction. Setting `output_chunk_length=1` and `output_chunk_shift=6` in your...