Training, Validation, and Testing

Training, validation, and test data support different stages of model development. A data split separates training data, validation data, and test data, helping prevent data leakage and produce a more credible final evaluation.

Japanese terms

1. Training, validation, and test data — 訓練くんれん・検証けんしょう・テストデータ: Separate data subsets used to fit a model, tune choices, and estimate final performance.
2. Data split — データ分割ぶんかつ: The process of partitioning a dataset into distinct subsets for model development and evaluation.
3. Training data — 訓練くんれんデータ: Examples used to fit a model’s parameters or learned behavior.
4. Validation data — 検証けんしょうデータ: Data used during development to compare settings or select models without fitting their parameters directly.
5. Test data — テストデータ: Held-out data used to estimate a finished model’s performance on unseen examples.
6. Data leakage — データリーク: The unintended use of information during training that would not be available when the model is deployed.

Related topics

Model evaluation applies metrics and validation procedures to these data subsets.