.rst .pdf

Quality Metrics

In oneDAL, a quality metric is a numerical characteristic or a set of connected numerical characteristics that represents the qualitative aspect of the result returned by an algorithm: a computed statistical estimate, model, or result of decision making.

A common set of quality metrics can be defined for some training and prediction algorithms.

A typical workflow with quality metric set is the following:

1. Create a quality metric set object to compute quality metrics.

   • Set specific parameters for the algorithms.

   • Use the useDefaultMetrics flag to specify whether the default or user-defined quality metrics should be computed.

2. Get an input collection object using QualityMetricsId of a specific algorithm.

3. Set data to the input collection using the algorithm’s InputId.

4. Perform computation.

5. Get the resulting collection of quality metrics using the algorithm’s ResultId.

Note

For values of InputId, Parameters, QualityMetricsId, ResultId, refer to the description of a specific algorithm.

Quality metrics are optional. They are computed when the computation is explicitly requested.

• Working with the Default Metric Set
  • Quality Metrics for Binary Classification Algorithms
  • Quality Metrics for Multi-class Classification Algorithms
  • Quality Metrics for Linear Regression
  • Quality Metrics for Principal Components Analysis
• Working with User-defined Quality Metrics

previous

Quantile

next

Working with the Default Metric Set