Running an Ablation Study

Ablation studies in PyKEEN.

An ablation study is an experiment in which components of a machine learning system are removed/replaced in order to measure the impact of these components on the system’s performance. In PyKEEN, a user could for instance measure the impact of explicitly modeling inverse relations on the model’s performance. This can be done with the pykeen.ablation.ablation_pipeline() function.

from pykeen.ablation import ablation_pipeline

ablation_result = ablation_pipeline(
    datasets='kinships',
    models=['RotatE', 'TransE'],
    losses=['BCEAfterSigmoidLoss', 'NSSA'],
    optimizers='Adam',
    training_loops=['sLCWA', 'LCWA'],
    optuna_config={
        'n_trials': 5,
    },
    directory='~/Desktop/simple_ablation_study',
)

Note

This tutorial will be improved with pykeen/pykeen#116.