opendataval.dataval.RobustVolumeShapley

class opendataval.dataval.RobustVolumeShapley(*args, **kwargs)

Robust Volume Shapley and Volume Shapley data valuation implementation.

While the following DataEvaluator uses the same TMC-Shapley algorithm used by semivalue evaluators, the following implementation defaults to the non-GR statistic implementation. Instead a fixed number of samples is taken, which is closer to the original implementation here: https://github.com/ZhaoxuanWu/VolumeBased-DataValuation/tree/main

References

[1]

X. Xu, Z. Wu, C. S. Foo, and B. Kian, Validation Free and Replication Robust Volume-based Data Valuation, Advances in Neural Information Processing Systems, vol. 34, pp. 10837-10848, Dec. 2021.

Parameters

samplerSampler, optional

Sampler used to compute the marginal contributions. Can be found in sampler, by default uses *args, **kwargs for GrTMCSampler.

robustbool, optional

If the robust volume measure will be used which trades off a “more refined representation of diversity for greater robustness to replication”, by default True

omegaOptional[float], optional

Width/discretization coefficient for x_train to be split into a set of d-cubes, required if robust is True, by default 0.05

Mixins

ModelLessMixin

Mixin for a data evaluator that doesn’t require a model or evaluation metric.

__init__(sampler: Sampler | None = None, robust: bool = True, omega: float | None = None, *args, **kwargs)

Methods

__init__([sampler, robust, omega])
embeddings(*tensors)	Returns Embeddings for the input tensors
evaluate_data_values()	Return data values for each training data point.
input_data(x_train, y_train, x_valid, y_valid)	Store and transform input data for volume-based evaluators.
input_fetcher(fetcher)	Input data from a DataFetcher object.
setup(fetcher[, pred_model, metric])	Inputs model, metric and data into Data Evaluator.
train(fetcher[, pred_model, metric])	Store and transform data, then train model to predict data values.
train_data_values(*args, **kwargs)	Trains model to predict data values.

Attributes

Evaluators
data_values	Cached data values.