def main(dataset_name, s):
print 'Reading dataset: ', dataset_name
num_class, num_feature, x_train, y_train, x_test, y_test = read_dataset(dataset_name)
x = np.vstack((x_train, x_test))
y_train.shape = (len(y_train), 1)
y_test.shape = (len(y_test), 1)
y = np.vstack((y_train, y_test))
y = y[:, 0]
# plot the first 2 dimensions
#plot_subspace(x[:, :2], y, num_class)
if s > num_feature:
print 'The target dimension must be equal or smaller than the original dim.'
eig_values, eig_vectors = FisherLDA(num_class, num_feature, x_train, y_train)
xt = np.matrix(x) * eig_vectors[:, :s]
plot_subspace(xt, y, num_class)
def main(dataset_name):
print 'Reading dataset: ', dataset_name
num_class, num_feature, x_train, y_train, x_test, y_test = read_dataset(dataset_name)
# PCA
#eig_values, eig_vectors = PCA(x_train, num_feature)
#DR_method = 'PCA'
# FisherLDA
eig_values, eig_vectors = FisherLDA(num_class, num_feature, x_train, y_train)
DR_method = 'FisherLDA'
rng = range(5, num_feature, 5)
rng.append(1)
for s in [ 64]:
# reduce the dimension of training set and test set
x_train_dr = np.matrix(x_train) * eig_vectors[:, :s]
x_test_dr = np.matrix(x_test) * eig_vectors[:, :s]
x_train_dr = np.array(x_train_dr)
x_test_dr = np.array(x_test_dr)
y_pred = classify_QDF(num_class, s, x_train_dr, y_train, x_test_dr)
classifier = 'QDF'
print '%s and %s reports with dim: %d, accuracy: %f' % (DR_method, classifier, s, sklearn.metrics.accuracy_score(y_test, y_pred))
y_pred = classify_LDF(num_class, s, x_train_dr, y_train, x_test_dr)
classifier = 'LDF'
print '%s and %s reports with dim: %d, accuracy: %f' % (DR_method, classifier, s, sklearn.metrics.accuracy_score(y_test, y_pred))
y_pred = classify_MQDF(num_class, s, x_train_dr, y_train, x_test_dr)
classifier = 'MQDF'
print '%s and %s reports with dim: %d, accuracy: %f' % (DR_method, classifier, s, sklearn.metrics.accuracy_score(y_test, y_pred))
print ''
f.close()
y_pred = classify(mean, sigma2, weight, x_train)
print 'Training acc:', sklearn.metrics.accuracy_score(y_train, y_pred)
y_pred = classify(mean, sigma2, weight, x_test)
print 'Test acc:', sklearn.metrics.accuracy_score(y_test, y_pred)
if __name__ == '__main__':
""" Load dataset and start to train the network
"""
import sys
dataset_name = sys.argv[1]
num_hidden = int(sys.argv[2]) # number of hidden nodes
print 'Reading dataset: ', dataset_name
num_class, num_feature, x_train, y_train, x_test, y_test = read_dataset(dataset_name, scaled=True)
# set a validation set to check convergence
count = len(x_train) / 3
# Note: the dataset has been shuffled
x_validation = np.array(x_train[:count])
y_validation = np.array(y_train[:count])
x_train = x_train[count:]
y_train = y_train[count:]
main()