Transformer-Hawkes-Process icon indicating copy to clipboard operation
Transformer-Hawkes-Process copied to clipboard
verified publisher icon SimiaoZuo

I wonder why the paper and code results are so different.

Open Yoontae6719 opened this issue 4 years ago • 4 comments

I ran the code as shown in your run.sh. As a result, it was confirmed that the RMSE was over 100, and the paper states that it is 4.99. Why is this?

Yoontae6719 avatar Nov 03 '21 06:11 Yoontae6719

The scale of the reported results is different from the scale of timestamp in the dataset.

ritvik06 avatar Nov 06 '21 19:11 ritvik06

@ritvik06 . As I know, this paper refered to the tables in the paper "self-attentive hawkes process" by quiang zhang and I saw the released code of Self-attentive hawkes process did not scale the timestamp. If the author changes the scale of timestamp, it is problematic because the scale of timestamp can reduce the negative log-liklihood due to the intergral term and RMSE easily. I cannot still understand why the author did not present the scaling parameter, and different reproducing code.

KanghoonYoon avatar Jun 10 '22 06:06 KanghoonYoon

I also find this question,and I dont know how to get the scale of timestamp, I hope someone can help me, thanks a lot!

TendaG0 avatar Oct 09 '22 06:10 TendaG0

anyone can explain to me, please?

TendaG0 avatar Nov 02 '22 03:11 TendaG0