Can you please describe how you setup the Long horizontal link forecasting evaluation

[JuliaGast]

in the paper you describe (e.g. in Figure 4: Long horizontal link forecasting: time-aware filtered MRR (%) on ICEWS05-15 with regard to different ∆t.)

Can you please describe how you did the experiment?
Did you write an additional script to parse the args args.delta_step= [0,1,2,3,4,....,8] subsequently, and for each delta store the resulting MRR?

Or, did you modify the code to look into the future at multiple target steps, and then set args.target_step=8?

Or anything I've missed?

Thanks!