def gen_Classifiers_name()

in main.py [0:0]

• 32 lines of code
• 6 McCabe index (conditional complexity)

def gen_Classifiers_name(args, transformer, datasets):
    """Generator of uninstantiated classifiers."""
    gamma = args.clfs.gamma_force_generalization

    data_weight = len(datasets["train"]) / len(datasets["test"])

    for n_hid in OmegaConf.to_container(args.clfs.nhiddens, resolve=True):
        for n_lay in OmegaConf.to_container(args.clfs.nlayers, resolve=True):
            for k_pru in OmegaConf.to_container(args.clfs.kprune, resolve=True):
                clf_name = (
                    f"clf_nhid_{n_hid}/clf_nlay_{n_lay}/clf_kpru_{k_pru}/gamma_{gamma}/"
                )

                Classifier = partial(
                    MLP, hidden_size=n_hid, n_hidden_layers=n_lay, k_prune=k_pru
                )

                kwargs = {}
                if not args.clfs.is_reinitialize:
                    kwargs["previous_mlp"] = transformer.module_.Q_zy

                Classifier = partial(
                    NeuralNetClassifier,
                    module=partial(
                        MCTrnsfClassifier,
                        transformer=transformer.module_,
                        Classifier=Classifier,
                        **OmegaConf.to_container(args.clfs.kwargs, resolve=True),
                        **kwargs,
                    ),
                    # don't use any regularization if you only care about training (e.g. Rademacher)
                    optimizer=get_optim(args),
                    criterion=partial(
                        CrossEntropyLossGeneralize,
                        gamma=gamma * data_weight,
                        map_target_position=datasets["train"].map_target_position,
                    ),
                )

                yield Classifier, clf_name