I found that "gpu" and "cpu" had a big impact on the results.

dreamquark-ai / tabnet

PyTorch implementation of TabNet paper : https://arxiv.org/pdf/1908.07442.pdf

https://dreamquark-ai.github.io/tabnet/

MIT License

2.56k stars 473 forks source link

Closed cc360 closed 2 years ago

cc360 commented 2 years ago

GPU：auc0.66 CPU：auc0.69

Is it because other parameters will change accordingly？

Why？thanks

eduardocarvp commented 2 years ago

Hello @cc360 ,

Switching from CPU to GPU does actually change some configurations. You can have more information here https://github.com/dreamquark-ai/tabnet/issues/345.

Let me know if this helps!