active_learning.py 文件源码

def run(train_data, test_data, n_features, labels, gamma, C, feature_type='uniform'):
    print train_data
    assert os.path.isfile(train_data), '%s is not a file' % train_data
    assert os.path.isfile(test_data), '%s is not a file' % test_data

    X, Y = load_svmlight_file(train_data, n_features=n_features)
    Xt, Yt = load_svmlight_file(test_data, n_features=n_features)
    Xt = Xt.todense()

    if gamma is None:
        gamma = 1.0 / n_features

    if C is None:
        C = 1

    rbf_svc = svm.SVC(kernel='rbf', gamma=gamma, C=C).fit(X, Y)

    print '--------------- original -----------------'
    baseline = sm.accuracy_score(Yt, rbf_svc.predict(Xt))
    print 'original: %f' % baseline

    CAL_v(train_data, labels[1], labels[0], rbf_svc.predict, n_features, feature_type, Xt, Yt)


# run('data/diabetes.aa', 'data/diabetes.ab', 8, (+1, -1), gamma=2.0, C=.5, feature_type='uniform')
# run('data/breast-cancer.aa', 'data/breast-cancer.ab', 10, (1, 0), gamma=0.5, C=.125, feature_type='uniform')
# run('data/australian.aa', 'data/australian.ab', 14, gamma=0.03125, C=.125, feature_type='uniform')
# run('./data/fourclass.aa', './data/fourclass.ab', 2, (1, -1), gamma=8.0, C=128, feature_type='uniform')