tasks.base_task¶
Tasks come above datasets in hierarchy level. In case you want to
implement a new task, you need to inherit BaseTask class. You need
to implement _get_available_datasets and _preprocess_item functions
to complete the implementation. You can check the source to see if you need
to override any other methods like prepare_batch.
Check example of VQATask here.
Example:
from pythia.tasks.base_task import BaseTask
from pythia.common.registry import registry
@registry.register_task("my")
class MyTask(BaseTask):
def __init__(self):
super().__init__("my")
def _get_available_datasets(self):
return ["my"]
def _preprocess_item(self):
item.text = None
return item
-
class
pythia.tasks.base_task.BaseTask(task_name)[source]¶ BaseTask that task classes need to inherit in order to create a new task.
Users must implement
_get_available_datasetsand_preprocess_itemin order to complete implementation.Parameters: task_name (str) – Name of the task with which it will be registered -
_get_available_datasets()[source]¶ Set available datasets for this task here. Override in your child task class Temporary solution, later we will use decorators to easily register datasets with a task
Returns: List - List of available datasets for this particular task
-
_init_args(parser)[source]¶ Override this function to add extra parameters to parser in your child task class.
Parameters: parser (ArgumentParser) – Original parser object passed from the higher level classes like trainer Returns: Description of returned object. Return type: type
-
_preprocess_item(item)[source]¶ Preprocess an item to be returned from __getitem__. Override in your child task class, so you have control on what you are returning
Parameters: item (Sample) – Sample returned by a particular dataset Returns: Preprocessed item Return type: Sample
-