train on my own dataset

[zhw2020913]

In the code, I use features of different dimensions, and the model will report a dimension error. I was wondering if I could change the model so that the model not only supports the unified 320 features