Customize datasets tutorial cant work about semi-supervised-learning

[cgxjdzz]

I totally understand after reviewing the code. There are some methods like train or evaluate in class AlgorithmBase , which wont be used in the whole process. And the dataset and dataloader created in the algorithm won't affect model training or predicting. I suppose you will delete these parameters and methods in the next version.

[Hhhhhhao]

Hi, I have fixed the colab for custom dataset. The reason it fails is due to 'none' in set for dataset in the config dict, which is not supported for not creating a dataset yet.

[cgxjdzz]

Thanks a lot! So this parameter has no influence on the following process?

[Hhhhhhao]

It only controls which dataset to be created. And for semilearn package we decide to re-create the dataset and data loader outside the trainer.

The 'none' option will be supported in next update to avoid the re-creation.

[dddZhao]

Hi, I still met this problem when I run the code locally. How to solve it?