search

bigdata-ustc / XKT

Multiple Knowledge Tracing models implemented by mxnet
MIT License
19 stars 5 forks source link

SKT:Training problem of SKT model #15

Open Dajcaac opened 2 years ago

Dajcaac commented 2 years ago

Hello, excuse me. I encountered some problems in the process of reproducing SKT. When I used the Assistment2015 data set, the learning rate and batch_size were the same as in the paper (lr=0.001, bs=16), but the AUC(macro_auc) of the training result was only 0.62. Is there anything I should pay attention to during training? Why my AUC results are so bad. Also, could you please explain the meaning of "macro_aupoc"? I don't quite understand what it means. Finally, I look forward to your reply. Thank you very much

XDZxdz1 commented 2 years ago

Hello, I also want to reproduce SKT, but when I reproduce the prerequisite graph and similar graph of assist2015, I cannot get the number of two edges reported in the paper. So I want to ask you, how do you reproduce these two kinds of graphs? Can you teach me

Dajcaac commented 2 years ago

Hello, I also want to reproduce SKT, but when I reproduce the prerequisite graph and similar graph of assist2015, I cannot get the number of two edges reported in the paper. So I want to ask you, how do you reproduce these two kinds of graphs? Can you teach me

I'm afraid I can't help you, because the number of edges I get is not the same as in the paper

XDZxdz1 commented 2 years ago

I also want to ask you whether this SKT code can only run on the CPU? What should I do if I want to transfer to GPU for running? macro_aupoc: I see its calculation method in the code. The path is XKT-master\Lib\sitepackages\longling\ML\metrics\classification.py

Dajcaac commented 2 years ago

I also want to ask you whether this SKT code can only run on the CPU? What should I do if I want to transfer to GPU for running? macro_aupoc: I see its calculation method in the code. The path is XKT-master\Lib\sitepackages\longling\ML\metrics\classification.py

You can download the GPU version of mxnet and change the ctx parameter to gpu()