search

CIA-Oceanix / TrAISformer

Pytorch implementation of TrAISformer---A generative transformer for AIS trajectory prediction (https://arxiv.org/abs/2109.03958).
Other
108 stars 37 forks source link

the result #1

Closed CdNut closed 2 years ago

CdNut commented 2 years ago

Professor Duong Nguyen, Thank you very much for sharing the code, but using the default config I can't reproduce the results in the paper, I can just get (0.8647, 1.6386, 2.7692) for (1 hour, 2 hours, 3 hours), and the train loss will reduce to negative number, I can not understand the reason. Could you please give me some advice?

dnguyengithub commented 2 years ago

Hello,

Please note that the values given by the code are in km, while the values presented in the paper were converted to nautical mile. It seems to me that you got the correct results. We used cross-entropy loss, it's normal that the training loss is negative.