Result on 1k

[seominjoon]

Would you be able to report the result on the 1k dataset as well? DMN+ paper doesn't report the 1k result so I was always wondering.. Thanks for great work!

[therne]

I haven't tested yet because it need extra hyperparameter optimization to prevent overfitting on smaller data - unlike the original DMN 😢 but I'll try and upload it soon!

[tyler7585]

Would you please upload your training model? I just download your code and follow the instructions , but no matter how I tried to train, the accuracy is much lower than what you said. I guess it might be some problem.

[therne]

@tyler7585 Unfortunately I've deleted all of my trained models 😢 and It'll take much time to train again and upload the entire models.

Can you tell me the task numbers that you're in trouble? It'll be much faster to upload those tasks only. (personally #19 was hard to reach that result - trained about 15 times and achieved it)

[tyler7585]

Would you please just tell me how much times it will take to train for the first task, the accuracy just fluctuate between 19~21. I have trained it for 2 times(256 epoch per time). So I thought maybe there is a problem, and do not train it any more.

[tyler7585]

[tyler7585]

BTW, I am training with your default setting.

[tyler7585]

@therne Is this situation normal? Because I don't find it has a clear increase.

[tomsonsgs]

@tyler7585 do you solve the problem?i meet the same problem