[API] Design approach for data metrics dependency about lightning-transformers HOT 2 CLOSED

class LitAutoModelTextClassificationTransformer(LitAutoModelTransformer):
    @property
    def num_classes():
        return self.trainer.datamodule.num_classes
    def self.configure_metrics():
        self.metric1 = metric(self.num_classes)
    def on_fit_start(self):
        self.configure_metrics()

class SomeoneElsesDataModule(LitTransformerDataModule):
    @property
    def C():
        # this user chose to set num_classes as C
        ...

# If necessary, one could create this "adapter" without rewriting configure_metrics
class AdaptedTransformer(LitAutoModelTextClassificationTransformer):
    @property
    def num_classes():
        return self.trainer.datamodule.C