Can anyone please specify the stride of the windows ?
For example if the first window goes from t1, t2, ..., t360
then does the second window goes from t2, t3, ...., t361
or it goes from t361, ...., t720
Hi,
The windows do not coincide, and they are not in chronological order.
MAFAT Challenge Team
Is the data within each window in chronological order and always separated by 0.5 seconds gap?
Posted by: devnikhilmishra @ May 19, 2022, 4:43 a.m.Yes, indeed.
MAFAT Challenge Team
I am wondering when you have split the windows, do you shuffle them? does it already happen in the public test set?
Otherwise, people might come up with some solution such as RNNs to "implicitly" keep track of previous states and use them to predict the current observation.
Note that, in case we have the full time series (of 5 hours), it is not too difficult to predict which segments the room is empty.
I think you asked this question https://codalab.lisn.upsaclay.fr/forums/3498/477/. We are not allowed to use the state or results from previously seen timesteps to predict the current one
Posted by: devnikhilmishra @ May 21, 2022, 8:31 a.m.@devnikhilmishra: In that thread, the organizer only said we are not allowed to keep previous-seen data.
Not sure about RNNs, why does the test set have some "metadata" ? Could we exploit them?
Posted by: tak @ May 21, 2022, 8:42 a.m.Hi,
Please refer to section 17 of the competition rules:
"The test set will include three-minute segments, each segment includes RSSI values referring to one pair of router-device. The prediction should be done by a single, individual segment, independently between segments."
Please make sure that your model fulfills the
conditions of the competition.
Notice that the windows in the private test set will be arranged in a random (not chronological) order.
MAFAT Challenge Team
Posted by: MAFAT_Challenge @ May 23, 2022, 4:18 a.m.