Spatio-Temporal Beam-Level Traffic Forecasting Challenge by ITU
'%3e%3cpath fill='%23001489' d='M0 0v6h9V0z'/%3e%3cpath fill='%23e03c31' d='M0 0v3h9V0z'/%3e%3cg stroke='%23fff' stroke-width='2'%3e%3cpath id='d' d='m0 0 4.5 3L0 6m4.5-3H9'/%3e%3cuse xlink:href='%23b' stroke='%23ffb81c' clip-path='url(%23c)'/%3e%3c/g%3e%3cuse xlink:href='%23d' fill='none' stroke='%23007749' stroke-width='1.2'/%3e%3c/g%3e%3c/svg%3e){kind=small}
Well ... the end is near ...
Test here really consists of two separate segments, the 6th week and the 11th week. Presumably these will be scored with equal weight in private and that can result in a big shakeup. The 11th week in particular is in the distant future and if your model e.g. forecasts an increase while in test there is a decrease, you can see a dramatic deterioration in your score. So the 11th week is quite important ...
Because of that, at one stage, I modelled the 6th and 11th week differently. Now I do not really want to discuss my model (just yet) but, during that time, I noted that it appears as if the 11th week is not scored in the public test! I have yet to test this, and given the constraint on subs probably won't, but if it is indeed the case, that the 11th week will only be scored in private, this competition may perhaps have a huge shake-up.
Is there any test data from the 2nd target week in public test - Zindi
Ok, did not see that but then, yes!, shake-up!
Fingers Crossed. It will be interesting to see different validation approaches :)
But this competition surely led me to explore all sorts of time series modeling techniques.
Well, the LB is heating up now as things draw to a close ... at least your model seems to be holding up very well.
Yes, as we go into the final lap, best wishes to you also! Let's wrap this up, really, time to move on to something new also ...
Data from the 11th week is definitely not in the public LB. I had a bug in my submission code where I accidentally copied week6 data to week11 and the score was the same before and after the fix.
I don't have a good model for this , I think you will be winner couse you did many experiments good luck and i think the private score will be greater than .23
looks like private rank is showing already.