Questions about Event Discovery Algorithm

[kibbon]

Hello, I've read your paper "Deep Sequence Learning with Auxiliary Information for Traffic Prediction" and find that the code of Event Discovery Algorithm may be not in accordance with its describtion in paper.
@ /src/preprocessing/new_anomalty1109.py line 38-43
The code shows that rules to discover event is:
(d(x,y,t) - d(x,y,t-7d))) > 300 && (d(x,y,t) / d(x,y,t-7d))) > 0.2 (The second condition make no sense)
In paper this rule to discover event is :
(d(x,y,t) - d(x,y,t-7d))) > 300 && (d(x,y,t) - d(x,y,t-7d))/d(x,y,t-7d) > 0.2

[bbliao]

@kibbon Yes, there is discordance between the paper and code. I am sorry for the carelessness since when my partner showed me the result, everything seems ok, then I did not review the code carefully. The difference is that the rule in the code will discover a little more events. (The "events" discovered by the code and their query counts are more than 300/0.2=1500 may not be "events" in the paper.)

So, If you just want to use the event discovery algorithm for event discovery, I suggest you use the rule in the paper. If you want to use our dataset and compare with our results, I suggest you use the code we provided.

I am sorry for the discordance again! Thanks for your attention!