Nixtla's Open Source Time Series Ecosystem.

Nixtla Community

~I've run a NeuralForecast object with four models: `AutoLSTM`, `AutoNHITS`, `AutoNBEATSx` and `AutoTFT` all using a loss function of `MQLoss(levels=[75, 80, 90, 95, 99])`. `AutoLSTM`, `AutoNHITS` produced the right columns, but `AutoNBEATSx` and `AutoTFT` produced columns and assumedly predictions for the following levels: [14900, 15900, 17900, 18900, 19700], but just the highs and not the lows. What's causing the issue? Are there only some models compatible with `MQLoss`?~

Looks like I had used quantiles instead of levels for the latter models.