Haozhe Zhu

参考文献部分 全部被标红了 T. T 有啥解决方法么？

> > 参考文献部分 全部被标红了 T. T > > 有啥解决方法么？ > > 啊这……最好联系一下学校（或者查重方）那边吧，一般来说参考文献是应该被排除掉的。 已经解决了。准确的说，是文中所有中文都变成了乱码，因此“参考文献”四个字没有被捕捉到，导致后续英文全部被查重。 这个问题是因为我用macOS Preview编辑了生成的PDF，凡是它保存的PDF都无法从PDF转回Word，因此知网无法识别其内容。这应该是macOS的问题（也可能是个feature），但反正和fdu-thesis这个项目没有关系。

Can this be implemented as a plugin?

For now, I have not changed the style. (Just add an option for ICP info) Let's move to #88 to continue discussion.

Maybe it would be better to add an option to allow users hide the `Powered by Hexo & Minos` information. (as the theme NeXT does.)

 And the hyberlink of the ICP information is a little eye-catching. I guess you can set it keeping in the same color with other ordinary text (i.e., in grey).

The bug of random_split has been discussed many times in the community, such as [Link](https://discuss.pytorch.org/t/how-to-split-test-and-train-data-keeping-equal-proportions-of-each-class/21063/10). It may produce an inconspicuous skewed val_dataset especially when the dataset is not large enough....

> the case why I have moved to PT split was to drop dependency on `sklearn` as this seems to be the only usage... :] > so shall we use...

Fixed in commit https://github.com/zhutmost/lsq-net/commit/02f802e6652d3d067b953f756f01dd2d50d54456.

However, the imbalance among categories still exists. Maybe need another `Sampler`.