filling dates with next day of forecast
Hi again @dbowdalo this is quite important for evaluating experiments that contain more than one day forecasts.
For example I was comparing the results using a7ej3 (that has 5day forecasts) and a7ej4 (contains only the first day forecast). Normally for this experiment (it's operational ICAP) there were many dates in summer that the runs were stopped (mn5 maintenance, overheating) so dates should be missing. I noticed that running reports for either a7ej3 and a7ej4 the timeseries look different: For a7ej3:
and for a7ej4:
This is important because if we have many dates that might be missing Providentia is using the second,third, fourth, fifth day of forecast that might be worse than the first and the stats will look different.