search

Xiaohan-Chen / TFPred

An official code for paper: TFPred: Learning discriminative representations from unlabeled data for few-label rotating machinery fault diagnosis
MIT License
44 stars 3 forks source link

结果问题 #4

Open nulloption opened 4 months ago

nulloption commented 4 months ago

请问作者，这里的实验结果并非像论文中所给的 b36514258a540c3eab8cf09d5c58087 论文中所给的上下误差不超过2%，但是我尝试了，发现结果上下浮动很大，可能超过10%

Xiaohan-Chen commented 3 months ago

Thanks for your interest. We have re-evaluated our method and found that the results are still the same with our paper. For validation, we have uploaded our pre-trained models in the history folder, feel free to test them.

As you mentioned the variance over 10, we did not observe this result, even in the compared methods (except baseline supervised). It seems that your model is underfitting, please check your hyperparameters carefully.