Method __init__

cdslib/core/script/base_train.py:61–251

Args: exp_tag (str): name of the experiment config_filename (str): the yaml or json filename of the model and dataloader, etc. trainer_filename (str): the previous trainer file to resume from. None: no resum

(
            self,
            exp_tag: str = "exp",
            config_filename: str = None,
            trainer_filename: str = None,
            work_dir: str = ".",
            output_dir: str = None,
            rank: int = 0,
            n_gpus: int = 1,
            random_seed: int = 0,
            save_code: bool = True,
            exclude_dirs: T.List[str] = None,
            exp_tag_first: bool = False,
            open_tensorboard: bool = True,
            ddp_type: str = "ddp",
            start_epoch: int = 0,
            end_epoch: int = 1000000,
            max_train_epoch_batches: int = -1,
            max_valid_epoch_batches: int = -1,
            max_test_epoch_batches: int = -1,
            save_every_num_epoch: int = 1,
            validate_every_num_epoch: int = 1,
            test_every_num_epoch: int = 1,
            log_every_num_train_batch: int = 1,
            log_every_num_valid_batch: int = 1,
            log_every_num_test_batch: int = 1,
            visualize_every_num_train_batch: int = 1,
            visualize_every_num_valid_batch: int = 1,
            visualize_every_num_test_batch: int = 1,
            tensorboard_exe_path: str = "tensorboard",
            # overwrite_pretrained_names: T.List[str] = None,
            use_torchrun: bool = True,
            find_unused_parameters: bool = False,  # used for DDP, see _setup_for_distributed_learning
            *args,
            **kwargs,
    )

    """

    def __init__(
            self,
            exp_tag: str = "exp",
            config_filename: str = None,
            trainer_filename: str = None,
            work_dir: str = ".",
            output_dir: str = None,
            rank: int = 0,
            n_gpus: int = 1,
            random_seed: int = 0,
            save_code: bool = True,
            exclude_dirs: T.List[str] = None,
            exp_tag_first: bool = False,
            open_tensorboard: bool = True,
            ddp_type: str = "ddp",
            start_epoch: int = 0,
            end_epoch: int = 1000000,
            max_train_epoch_batches: int = -1,
            max_valid_epoch_batches: int = -1,
            max_test_epoch_batches: int = -1,
            save_every_num_epoch: int = 1,
            validate_every_num_epoch: int = 1,
            test_every_num_epoch: int = 1,
            log_every_num_train_batch: int = 1,
            log_every_num_valid_batch: int = 1,
            log_every_num_test_batch: int = 1,
            visualize_every_num_train_batch: int = 1,
            visualize_every_num_valid_batch: int = 1,
            visualize_every_num_test_batch: int = 1,
            tensorboard_exe_path: str = "tensorboard",
            # overwrite_pretrained_names: T.List[str] = None,
            use_torchrun: bool = True,
            find_unused_parameters: bool = False,  # used for DDP, see _setup_for_distributed_learning
            *args,
            **kwargs,
    ):
        """
        Args:
            exp_tag (str):
                name of the experiment
            config_filename (str):
                the yaml or json filename of the model and dataloader, etc.
            trainer_filename (str):
                the previous trainer file to resume from. None: no resume.
            output_dir (str):
                the root dir of the outputs
            rank (int):
                the rank of the process. Optional if use torchrun to launch.
            n_gpus (int):
                number of gpus (world size). Optional if use torchrun to launch.
            random_seed (int):
                random seed
            save_code (bool):
                whether to save the code in output_dir for future reference.
            exclude_dirs (list of str):
                dirs to exclude from saving code
            exp_tag_first (bool):
                whether the output_dir struture is: