Skip to main content
Version: Next

Metrics: Size dependency

Overview​

H2O Model Validation offers an array of metrics to understand a size dependency test. Below, each metric is described in turn.

Settings​

Modeling scores for different train dataset sizes (MSDTDS)​

The MSDTDS graph visualizes several scorer values for models (identical models) trained with different training data sizes. Observing the MSDTDS plot graph can enable you to understand and find an acceptable train dataset size for your model.

  • Y-axis: Scorer values
  • X-axis: Size of the training data
  • Validation: The validation scores of each model
  • Test: The test scores of each model
  • Dots: Dots on each line refer to a child model or the parent model
    • Child model: A child model refers to a model trained on one of the sub-training samples obtained from the original training data.
    • Parent model: The parent model refers to a trained model with the original training data and not with one of the sub-training samples.

Scores for models in the MSDTDS plot are positioned from left to right. The first model on the left (also known as a child model) represents the model with the smallest sub-training sample obtained from the original train data. The model furthest to the right, known as the parent model, represents the original model that used the whole original train data. The model right before the parent model represents the model with the highest sub-training sample.

MSDTDS

Feature importance for different training data sizes (FIDTDS)​

The FIDTDS heatmap visualizes the most important features for different models trained on different sizes of training data during the retraining process of a model.

  • Rows: Dataset variables (features)
  • Columns: Train dataset size

FIDTDS

Models tab​

Models table​

The models table displays the H2O Driverless AI experiments corresponding to size dependency models at each split.

Column nameDescription
#This column defines the experiment number.
Model originThis column defines whether the model is the parent model or a child model.
EnsembleThis column defines the ensemble of models used in the Driverless AI experiment with their weights.
Best modelThis column defines the best machine learning (ML) model used in the Driverless AI experiment.
Train dataset sizeThis column defines the size of the training data used in the Driverless AI experiment.
Best featureThis column defines the best feature information of the Driverless AI experiment.
Feedback