EvaluationMetric

This class represents the available evaluation metrics for training the GNN models.

Parameters

Name	Type	Description	Optional
name	str	The name of the evaluation metric to use. Supported metrics vary depending on the task; see the table below for all available options.	No
eval_at_k	int	The number of top predictions (k) to consider when computing the evaluation metric; applicable only for link prediction tasks.	Yes

Supported Metrics

Metric Name	Task	Documentation Link
average_precision	binary_classification	link
accuracy	binary_classification or multiclass_classification	link
f1	binary_classification	link
roc_auc	binary_classification	link
precision	binary_classification	link
recall	binary_classification	link
multilabel_auprc_micro	multilabel_classification	link
multilabel_auroc_micro	multilabel_classification	link
multilabel_precision_micro	multilabel_classification	link
multilabel_auprc_macro	multilabel_classification	link
multilabel_auroc_macro	multilabel_classification	link
multilabel_precision_macro	multilabel_classification	link
macro_f1	multiclass_classification	link
micro_f1	multiclass_classification	link
r2	regression	link
mae	regression	link
rmse	regression	link
mape	regression	link
link_prediction_precision	link_prediction or repeated_link_prediction	link
link_prediction_recall	link_prediction or repeated_link_prediction	link
link_prediction_map	link_prediction or repeated_link_prediction	link

Returns

An instance of the EvaluationMetric class.

Examples

For a binary classification task:

from relationalai_gnns import EvaluationMetric

binary_clf_metric = EvaluationMetric(name="accuracy")

For a link prediction task:

from relationalai_gnns import EvaluationMetric

link_pred_metric = EvaluationMetric(name="link_prediction_map", eval_at_k=12)