The evaluation class should be based on DataCotainer using add_predictions. Plotting should be based on PlotterClass. As evaluation is similar depending on task. Design a general Evaluation Class to avoid stupid re-implementations.

After Trainer performed training. For example

dset = trainer.get_eval_data()  # Make y_pred / y_true
eval = EvalClass(dset, {"primary": "classification", "energy": "regression"}, mode="hess")

eval(ev_data)