bnelearn.experiment.experiment module

This module defines an experiment. It includes logging and plotting since they can often be shared by specific experiments.

class bnelearn.experiment.experiment.Experiment(config: ExperimentConfig)

Bases: ABC

Abstract Class representing an experiment

action_size: int

bne_env: AuctionEnvironment

bne_utilities: Tensor

epoch: int

input_length: int

mechanism: Mechanism

n_models: int

observation_size: int

plot_xmax: float

plot_xmin: float

plot_ymax: float

plot_ymin: float

positive_output_point: Tensor

pretrain_transform(player_position: int) → callable

Some experiments need specific pretraining transformations. In most cases, pretraining to the truthful bid (i.e. the identity function) is sufficient.

Args:

player_position (:int:) the player for which the transformation is

requested.

Returns

(:callable:) pretraining transformation

run() → bool

Runs the experiment implemented by this class, i.e. all defined runs.

If a run fails for whatever reason, a warning will be raised and the next run will be triggered until all runs have completed/failed.

Returns:

success (bool): True if all runs ran successfully, false otherwise.

valuation_size: int