Validation Curve Results
Not calculatedInterpretation
Curve summary
| Parameter | Training | Validation | Gap | Relative validation | Diagnosis |
|---|
Copy-ready report
Calculator Inputs
Formula Used
A validation curve compares training and cross-validation performance across hyperparameter values. Each plotted point normally represents a mean score from repeated folds.
Training mean: μtrain(λ) = (1 / k) × Σ TrainingScoreᵢ(λ) Validation mean: μvalid(λ) = (1 / k) × Σ ValidationScoreᵢ(λ) Standard deviation: σ(λ) = √[Σ(Scoreᵢ(λ) − μ(λ))² / (k − 1)] Generalisation gap: G(λ) = |μtrain(λ) − μvalid(λ)| Best parameter for maximisation: λ* = arg max μvalid(λ) Best parameter for minimisation: λ* = arg min μvalid(λ)
How to Use
- Enter the model, hyperparameter, and evaluation metric names.
- Select whether higher or lower metric values are better.
- Choose a linear, logarithmic, or categorical parameter axis.
- Enter training and validation means for every tested value.
- Add standard deviations when cross-validation fold results are available.
- Adjust diagnostic thresholds and chart display options.
- Click calculate to identify the strongest validation setting.
- Review the curve, gaps, diagnoses, and downloadable report.
Example Data
| Gamma | Training mean | Training std | Validation mean | Validation std |
|---|---|---|---|---|
| 0.0001 | 0.56 | 0.018 | 0.54 | 0.025 |
| 0.001 | 0.71 | 0.016 | 0.68 | 0.022 |
| 0.01 | 0.89 | 0.012 | 0.84 | 0.018 |
| 0.1 | 0.97 | 0.008 | 0.91 | 0.014 |
| 1 | 0.995 | 0.003 | 0.87 | 0.020 |
| 10 | 1.00 | 0.000 | 0.78 | 0.030 |
Frequently Asked Questions
What is a validation curve?
It plots training and validation scores against one hyperparameter. The comparison reveals model capacity changes across tested settings.
How is the best parameter selected?
The calculator chooses the strongest validation mean. It follows the selected maximise or minimise objective.
What does a large generalisation gap mean?
A large gap often indicates overfitting. Training performance remains stronger than validation performance.
How does underfitting appear?
Training and validation scores are both weak. Their gap is often comparatively small.
Should the parameter axis be logarithmic?
Use a logarithmic axis for positive values spanning powers of ten. Regularisation strengths and learning rates commonly benefit.
What are standard-deviation bands?
They show score variability across folds or repeated evaluations. Narrow bands usually indicate more stable estimates.
Can loss values be plotted?
Yes. Select lower is better for loss, error, or deviance metrics.
Can categorical parameter values be used?
Yes. Choose the categorical axis for labels such as kernels, solvers, or penalty types.
Does the tool train a machine learning model?
No. It analyses scores you provide from experiments, cross-validation, or model-selection software.
What can be exported?
You can copy a text report, download CSV results, save a PDF, or export the Plotly chart as PNG.