Weight calibration

The survey_enhance.reweight module contains two classes which can be used to define a loss architecture and minimise it for a given survey.

class survey_enhance.reweight.CalibratedWeights(initial_weights: numpy.ndarray, dataset: policyengine_core.data.dataset.Dataset, loss_type: Type[survey_enhance.reweight.LossCategory], calibration_parameters: policyengine_core.parameters.parameter_node.ParameterNode)

Bases: object

calibrate(time_instant: str, epochs: int = 1000, min_loss: Optional[float] = None, learning_rate: float = 0.1, validation_split: float = 0.0, validation_blacklist: Optional[List[str]] = None, rotate_holdout_sets: bool = False, log_dir: Optional[str] = None, tensorboard_log_dir: Optional[str] = None, log_frequency: int = 15, verbose: bool = False) → numpy.ndarray

calibration_parameters: policyengine_core.parameters.parameter_node.ParameterNode

dataset: policyengine_core.data.dataset.Dataset

initial_weights: numpy.ndarray

loss_type: Type[survey_enhance.reweight.LossCategory]

class survey_enhance.reweight.LossCategory(dataset: policyengine_core.data.dataset.Dataset, calibration_parameters_at_instant: policyengine_core.parameters.parameter_node_at_instant.ParameterNodeAtInstant, instant: Optional[str] = None, calibration_parameters: Optional[policyengine_core.parameters.parameter_node.ParameterNode] = None, weight: Optional[float] = None, ancestor: Optional[survey_enhance.reweight.LossCategory] = None, static_dataset: Optional[bool] = None, comparison_white_list: Optional[List[str]] = None, comparison_black_list: Optional[List[str]] = None, name: Optional[str] = None, normalise: Optional[bool] = None, diagnostic: Optional[bool] = None)

Bases: torch.nn.modules.module.Module

A loss category is essentially a loss function, but contains a number of utilities for ease of programming, like decomposition into weighted and normalised subcategories, and logging.

collect_comparison_log() → pandas.core.frame.DataFrame

computation_tree(household_weights: torch.Tensor, dataset: policyengine_core.data.dataset.Dataset, filter_non_one: bool = True) → dict

create_holdout_sets(dataset: policyengine_core.data.dataset.Dataset, num_sets: int, num_weights: int, exclude_by_name: Optional[str] = None) → List[Tuple[policyengine_core.data.dataset.Dataset, policyengine_core.data.dataset.Dataset]]

diagnostic: bool = False

Whether to log the full tree of losses.

diagnostic_tree: Dict[str, float] = None

The tree of losses.

evaluate(household_weights: torch.Tensor, dataset: policyengine_core.data.dataset.Dataset, initial_run: bool = False) → torch.Tensor

forward(household_weights: torch.Tensor, dataset: policyengine_core.data.dataset.Dataset, initial_run: bool = False) → torch.Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

get_comparisons(dataset: policyengine_core.data.dataset.Dataset) → List[Tuple[str, float, torch.Tensor]]

normalise: bool = True

Whether to normalise the starting loss value to 1.

static_dataset = False

Whether the dataset is static, i.e. does not change between epochs.

subcategories: List[Type[LossCategory]] = []

The subcategories of this loss category.

weight: float = 1.0

The weight of this loss category in the total loss.