def test_classifier_cv(self, classifier, name):
"""
Performs 10-fold cross-validation on the given classifier.
@param name {string} Name of the classifier or run
@param classifier {sklearn Classifier} Classifier instance
"""
folds = self.load_folds()
if not folds:
logger.error("Folds is None?")
# Does not handle root nodes separately as the folds do not distinguish
# between these
# So what we do here: collect all results, i..e all csv rows for all
# folds, and print two files with 10 lines each
logger.info("test_classifier: %s", classifier)
foldNum = 0
for train_index, test_index in self.folds:
logger.debug("Current fold: train_index: %s", train_index)
logger.debug("Current fold: test_index: %s", test_index)
classifier.fit(self.trainData[train_index],
self.trainLabels[train_index])
pred = classifier.predict(self.trainData[test_index])
allEval = evaluate.printStatsCoarseInt(
self.trainLabels[test_index],
pred)
allEval = evaluate.ins(['Classifier', 'Mode', 'Fold'],
[name, 'CV', foldNum],
allEval)
self.evalData.append(allEval)
# mangle name to include other columns
self.get_feature_importances(name + '-CV-Fold-' + foldNum,
classifier)
foldNum += 1
def test_classifier_toClassify(self, classifier, name):
"""
Trains classifier on EnsembleLearner::trainData and
predicts labels for EnsembleLearner::testData.
The predictions are saved as a tree to the location specified by
EnsembleLearner::toClassifyOutFile.
If EnsembleLearner::toClassifyIsTest is set, the labels
in EnsembleLearner::testLabels are used to evaluate the
predictions.
@param name {string} Name of the classifier or run
@param classifier {sklearn Classifier} Classifier instance
"""
logger.info("Started classifying file")
logger.debug("len(data): %s" % len(self.trainData))
logger.debug("len(labels): %s" % len(self.trainLabels))
classifier.fit(self.trainData, self.trainLabels)
self.get_feature_importances(name + '-Single', classifier)
pred = classifier.predict(self.testData)
out = open(self.toClassifyOutFile, "w")
trees = self.apply_prediction_to_tree(
self.toClassify, pred.tolist())
for tree in trees:
# .encode('utf-8'))
out.write(tree.pprint(margin=999999, nodesep='').encode('utf-8'))
out.write('\n')
out.close()
logger.info("Finished classifying file")
if self.toClassifyIsTest:
rootPred = numpy.asarray(pred)[self.testRootInd]
rootGold = numpy.asarray(self.testLabels)[self.testRootInd]
rootEval = evaluate.printStatsCoarseInt(rootGold,
rootPred,
"root")
allEval = evaluate.printStatsCoarseInt(self.testLabels,
pred)
allEval.update(rootEval)
allEval = evaluate.ins(['Classifier', 'Mode', 'Fold'],
[name, 'Single', 'N/A'],
allEval)
self.evalData.append(allEval)