Experimental results cannot reach the results of the paper

[MingjieWang0606]

Hello researcher, thank you very much for such an amazing paper and idea.
I am very amazed by this result and I would like to try to reproduce your result for this reason.
I set up a brand new conda environment and followed your experimental requirements to the letter.
But using python main.py --saved-model-name matchmaker.h5 --train-test-mode 1
I can only get a score of 88 for mse, which is very puzzling to me. Can you share your experimental environment and other tricks?

[hikuru]

Hi,

It could be due to early stopping since you are training the model. Can retry your training after increasing the earlyStop_patience value to 1000 in here

[MingjieWang0606]

Thank you for your reply!
I guessed before that this might be the problem. But the model trained 2000 epoch. but still can't reach the accuracy.
Also, I used a completely consistent environment.