graphstorm.eval

GraphStorm provides built-in evaluators and interfaces for different Graph Machine Learning tasks. Each evaluator can have multiple task specific metrics for model evaluation. For example, GSgnnClassificationEvaluator uses accuracy as its default evaluation metric. However, users can also set other metrics, e.g., precision_recall, roc_auc, and f1_score in GSgnnClassificationEvaluator.

Users can find the information about metrics for different tasks in the Evaluation Metrics section.

If users want to implement customized evaluators, a best practice is to extend the base evaluator, i.e., GSgnnBaseEvaluator, and the corresponding evaluation interfaces, e.g., GSgnnPredictionEvalInterface for prediction evaluation and GSgnnLPRankingEvalInterface for ranking-based link prediction evaluation, and then implement the abstract methods defined in those interface classes.

Base Evaluators

`GSgnnBaseEvaluator`	Base class for GraphStorm Evaluators.
`GSgnnPredictionEvalInterface`	Interface for Prediction evaluation functions.
`GSgnnLPRankingEvalInterface`	Interface for Link Prediction evaluation functions using ranking methods.

Evaluators

`GSgnnClassificationEvaluator`	Evaluator for classification tasks.
`GSgnnRegressionEvaluator`	Evaluator for regression tasks.
`GSgnnLPEvaluator`	Evaluator for Link Prediction tasks using “mrr” and/or "hit@k" as metric(s).
`GSgnnPerEtypeLPEvaluator`	Evaluator for Link Prediction tasks using `mrr` and/or `hit@k` as metric(s), and return per edge type scores.
`GSgnnRconstructFeatRegScoreEvaluator`	Evaluator for feature reconstruction tasks using regression scores.