def run()

in experiments/codes/experiment/checkpointable_multitask_experiment.py [0:0]

• 44 lines of code
• 18 McCabe index (conditional complexity)

    def run(self):
        """Method to run the experiment"""
        # experiment.run()
        if self.config.model.use_composition_fn:
            self.load_only_composition()
        if self.config.model.use_representation_fn:
            self.load_only_representation()
        if self.config.model.freeze_composition_fn:
            self.load_only_composition()
            self.composition_fn.freeze_weights()
        if self.config.model.freeze_representation_fn:
            self.representation_fn.freeze_weights()
        # re-register the params to the optimizer
        self.register_optim_sched(
            skip_composition_registry=self.config.model.freeze_composition_fn,
            skip_representation_registry=self.config.model.freeze_representation_fn,
        )
        if self.config.general.train_mode == "run_mult":
            self.run_multitask_training()
        elif self.config.general.train_mode == "run_mult_unique_comp":
            self.run_multitask_training_unique_composition()
        elif self.config.general.train_mode == "run_mult_unique_rep":
            self.run_multitask_training_unique_representation()
        elif self.config.general.train_mode == "supervised":
            self.run_single_task(world_mode="train")
        elif self.config.general.train_mode == "supervised_valid":
            self.run_single_task(world_mode="valid")
        elif self.config.general.train_mode == "supervised_test":
            self.run_single_task(world_mode="test")
        elif self.config.general.train_mode == "seq_mult":
            self.run_sequential_multitask_training()
        elif self.config.general.train_mode == "seq_mult_comp":
            self.run_sequential_multitask_unique_composition()
        elif self.config.general.train_mode == "seq_mult_rep":
            self.run_sequential_multitask_unique_representation()
        elif self.config.general.train_mode == "seq_zero":
            self.run_sequential_zeroshot_transfer()
        elif self.config.general.train_mode == "seq_full":
            self.run_sequential_fewshot_transfer(full_shot=True)
        elif self.config.general.train_mode == "seq_few":
            self.run_sequential_fewshot_transfer()
        elif self.config.general.train_mode == "pretrain":
            self.run_pretraining()
        else:
            raise NotImplementedError(
                "training mode not implemented. should be either one of \n supervised / seq_mult / seq_zero / seq_full / seq_few"
            )