I want to apply a LWR to a NDVI time series which is a union between estimates for Sentinel and Landsat, to smooth the values ​​and then make a daily interpolation.

I have hourly sales data (0600H – 2200H). I have aggregated them to daily and made predictions based on the daily level. This is because I found that predicting at daily level leads to predictions with higher accuracy.

I have been using Darts, where there is a very nice historical forecast functionality, and also nixtla’s neuralforecast library. The latter does have cross-validation, which can mimick a historical forecast, but is bog-slow (unlike the other parts of nixtla’s library, which are very speedy, and allow for a lot more customization than Darts). Any ideas of what the “right way” to do this is?