The following trains a classifier on a training set, predicts classes for the training and test data, and reports accuracy metrics.

$ classifyAndTest training_data.h5 test_data.h5 output_dir classifier_type subsample_param

See below for a details on the format of the expected h5 files.

One can subsample the input data during training to speed-up the process. The default behavior uses all the data for training ( subSample = "-1" ). Specifying a positive integer (N) will use that number of training samples. Specifying a float value in (0,1] will use that fraction of training samples.

This project accepts data in the HDF5 format

• The H5 file must have a 'labels' group with one Dataset per class.
• Currently classes must be named "1","2",..."N"
• The Dataset for each class must be 2d ( N x M )
  • N - number of observations
  • M - number of variables

The below is an example of the format for the famous iris data.

$ h5ls iris_all.h5
labels                   Group
$ h5ls iris_all.h5/labels
1                        Dataset {50, 4}
2                        Dataset {50, 4}
3                        Dataset {50, 4}

It is sometimes convenient to store data in multiple h5 files. In this case, one can specify a directory containing multiple h5 files. In this case, all h5 files in that directory will be included in the dataset.