Train/Validation/Test Split

Intermediate

Separating data into training (fit), validation (tune), and test (final estimate) to avoid leakage and optimism bias.

Full Definition

Separating data into training (fit), validation (tune), and test (final estimate) to avoid leakage and optimism bias.

Keywords

evaluation protocol

Domains

Evaluation & Benchmarking

Related Terms

PR Curve related to

Often more informative than ROC on imbalanced datasets; focuses on positive class performance.

 Brier Score related to

A proper scoring rule measuring squared error of predicted probabilities for binary outcomes.

 Experiment Tracking related to

Logging hyperparameters, code versions, data snapshots, and results to reproduce and compare experiments.

 Observability related to

A broader capability to infer internal system state from telemetry, crucial for AI services and agents.

Concept Map

See how Train/Validation/Test Split connects to other concepts.

Open Knowledge Graph