def process_arguments(self):
parser = argparse.ArgumentParser(description='Classifier meta-parameter optimization')
parser.add_argument('train', help='Train dataset')
parser.add_argument('test', help='Test dataset')
parser.add_argument('model', help='File to save best model')
parser.add_argument('scores', default=None, help='File to save scores of tested models')
parser.add_argument('-t', '--type', default='grid', choices=['grid', 'random', 'pso'], help='Search type')
parser.add_argument('-i', '--iterations', default=self.iterations, type=int, help='Iterations amount for pso and random search')
parser.add_argument('-j', '--jobs', default=-1, type=int, help='Processes amount for learning')
args = parser.parse_args()
trainData, trainLabel = loadDataset(args.train)
testData, testLabel = loadDataset(args.test)
self.initialize_optimizer(args.type, args.model, trainData, trainLabel, testData, testLabel, args.jobs, args.iterations, args.scores)
classifiers = [
('LogisticRegression', LogisticRegression),
# ('SVC', SVC),
# ('NuSVC', NuSVC),
('RandomForestClassifier', RandomForestClassifier),
]
for name, cl in classifiers:
c = cl()
print name
test_classifier(c, trainData, trainLabel, testData, testLabel)
def test_svc(trainData, trainLabel, testData, testLabel):
c = NuSVC(nu=0.05)
test_classifier(c, trainData, trainLabel, testData, testLabel)
if __name__ == '__main__':
import sys
if len(sys.argv) < 4:
print 'USAGE:\n\t' + sys.argv[0] + ' train.csv test.csv classifier.pkl'
sys.exit(1)
trainX, trainY = loadDataset(sys.argv[1])
testX, testY = loadDataset(sys.argv[2])
# test_svc(trainX, trainY, testX, testY)
# test_bunch_of_classifiers(trainX, trainY, testX, testY)
cl = RandomForestClassifier()
# cl = LogisticRegression()
cl = test_classifier(cl, trainX, trainY, testX, testY)
joblib.dump(cl, sys.argv[3])
r.fit(trainData, trainLabel)
print("Best parameters set found on development set:")
print(r.best_estimator_)
print("Grid scores on development set:")
for params, mean_score, scores in r.grid_scores_:
print("%0.3f (+/-%0.03f) for %r"
% (mean_score, scores.std() / 2, params))
return r
if __name__ == '__main__':
import sys
trainData, trainLabel = loadDataset(sys.argv[1])
testData, testLabel = loadDataset(sys.argv[2])
classifier = SVC()
param_grid = [
# {'C': [50, 300, 1000, 2500, 5000], 'kernel': ['linear']},
{'C': [1, 10, 100, 1000], 'gamma': [0.001, 0.0001], 'kernel': ['rbf']},
]
# gridSearch = gridSearch(classifier, param_grid, trainData, trainLabel)
# clf = gridSearch.best_estimator_
# rs = randomizedSearch(classifier, trainData, trainLabel)
# clf = rs.best_estimator_
po = pso_svc_optimization(trainData, trainLabel)
