/**
* Determine the list of test instances and the list of training instances.
*
* @param prefix
* @param type
* @param prune
*/
public void prepare(String prefix, SequenceType type) {
_classNames = Utils.getActivityNames(prefix);
_testMap = getTestSet(_classNames);
// build a training set and a test set.
_training = new HashMap<String, List<Instance>>();
_testing = new ArrayList<Instance>();
// Now train up the signatures...
for (String className : _classNames) {
File dataFile = new File("data/input/" + className + ".lisp");
List<Instance> instances = Utils.sequences(className, dataFile.getAbsolutePath(), type);
List<Integer> testSet = _testMap.get(className);
for (Instance instance : instances) {
if (testSet.contains(instance.id())) _testing.add(instance);
else {
List<Instance> list = _training.get(instance.name());
if (list == null) {
list = new ArrayList<Instance>();
_training.put(instance.name(), list);
}
list.add(instance);
}
}
}
}
/**
* Select a random set to be the test set.
*
* @param classNames
* @return
*/
public Map<String, List<Integer>> getTestSet(List<String> classNames) {
Random r = new Random(System.currentTimeMillis());
Map<String, List<Integer>> testSet = new HashMap<String, List<Integer>>();
for (String className : classNames) {
String f = "data/input/" + className + ".lisp";
Map<Integer, List<Interval>> map = Utils.load(new File(f));
List<Integer> episodes = new ArrayList<Integer>(map.keySet());
Collections.shuffle(episodes, r);
// 33% of the instances will be part of the test set
double pct = 1.0 / 3.0;
int number = (int) Math.round((double) episodes.size() * pct);
List<Integer> list = new ArrayList<Integer>();
for (int i = 0; i < number; ++i) {
list.add(episodes.get(i));
}
testSet.put(className, list);
}
return testSet;
}