PyHealth trainer uses deprecated model.mode attribute instead of checking its output or label processors

[jhnwu3]

PyHealth/pyhealth/trainer.py

Line 325 in 8d71772

if self.model.mode is not None:

Basically, we set a mode for what metrics we use, binary, multiclass, multilabel, etc.

The irony here is we can simply just directly map that mode with our output processors and thereby eliminating an intermediate variable.

This isn't a big deal now, but would be good to remedy as we start incorporating things like reconstruction losses, generative modeling techniques, etc. where we want to scale what metrics are compatible with not only the vanilla PyHealth trainer but also things like PyTorch lightning that is directly compatible with the litData SampleDataset generated by any set_task()