Infer.NET user guide : Learners : Bayes Point Machine classifiers : Command-line runners

Cross-validation

You can use the CrossValidate module to assess the generalization performance of the Bayes Point Machine, both in binary and multi-class classification. The CrossValidate module starts by reading a labelled data from a file and partitions its instances into K subsets of equal size, known as folds. It then trains the Bayes Point Machine classifier on K - 1 folds and evaluates its performance on the withheld K-th fold. It cycles through all K combinations of splits into training and validation sets to finally report the overall performance results.

The CrossValidate module has the following command-line arguments:

Required arguments

  • data-set: The file containing ground truth labels and features in the format described earlier.
  • results: The CSV file to which the cross-validation results will be saved.

Optional arguments

  • folds: The number of cross-validation folds to use (defaults to 5).
  • iterations: The number of training algorithm iterations (defaults to 30).
  • batches: The number of batches into which the training data is split (defaults to 1).
  • compute-evidence: If specified, the Bayes Point Machine classifier will compute model evidence on the training data (defaults to false).

For more information about the command-line arguments, see Settings.

Example

Learner Classifier BinaryBayesPointMachine CrossValidate 
    --data-set training.dat --results cross-validation-results.csv 
    --iterations 15 --batches 1 --compute-evidence

Learner Classifier MulticlassBayesPointMachine CrossValidate 
    --data-set training.dat --results cross-validation-results.csv 
    --iterations 15 --batches 1 --compute-evidence
©2009-2015 Microsoft Corporation. All rights reserved.  Terms of Use | Trademarks | Privacy Statement