"""
This module defines the `TrainingConfig` class, which is responsible for
managing and accessing training-related configurations from a YAML file.
It extends :class:`aiqclib.common.base.config_base.ConfigBase` to provide
structured access to dataset settings, including targets, step classes,
and step parameters, by resolving references within the configuration.
"""
from aiqclib.common.base.config_base import ConfigBase
from aiqclib.common.utils.config import get_config_item
[docs]
class TrainingConfig(ConfigBase):
"""
A configuration class providing interfaces for training dataset settings.
Inherits from :class:`ConfigBase` with an expectation of working
under the "training_sets" section in the YAML configuration.
Leverages methods like :meth:`select` to initialize and fetch
subset configurations (e.g., target sets, step parameters).
.. note::
:attr:`expected_class_name` must match the YAML's ``base_class``
property if you intend to instantiate this class directly from config.
"""
expected_class_name: str = "TrainingConfig"
"""
The class name expected by :class:`ConfigBase` for consistency checks
when instantiating TrainingConfig from YAML.
"""
def __init__(self, config_file: str, auto_select: bool = False) -> None:
"""
Initialize the training configuration.
:param config_file: The path to the YAML configuration file.
This file will be used to load the training sets.
:type config_file: str
:param auto_select: If True, automatically selects the first dataset
found in the `training_sets` section upon initialization.
:type auto_select: bool
:raises ValueError: If the YAML structure is invalid
or `training_sets` cannot be found.
"""
super().__init__(
section_name="training_sets",
config_file=config_file,
auto_select=auto_select,
)
[docs]
def select(self, dataset_name: str) -> None:
"""
Select a named dataset from the `training_sets` configuration,
retrieving nested configurations for targets, step classes,
and step parameters.
After calling :meth:`select`, sub-keys (``target_set``,
``step_class_set``, etc.) are populated from their respective
config dictionaries by resolving their references within the full configuration.
:param dataset_name: The key name of the dataset to select
within :attr:`data` (which references
the `training_sets` section).
:type dataset_name: str
:raises KeyError: If ``dataset_name`` is not found within
the `training_sets` dictionary.
"""
super().select(dataset_name)
self.data["target_set"] = get_config_item(
self.full_config, "target_sets", self.data["target_set"]
)
self.data["step_class_set"] = get_config_item(
self.full_config, "step_class_sets", self.data["step_class_set"]
)
self.data["step_param_set"] = get_config_item(
self.full_config, "step_param_sets", self.data["step_param_set"]
)