def neural_network(x_train, x_test, y_train, y_test, x_pca, x_ica, x_kpca,
x_rp, x_kmeans, x_gmm, **kwargs):
"""Perform neural network experiment.
Args:
x_train (ndarray): training data.
x_test (ndarray): test data.
y_train (ndarray): training labels.
y_test (ndarray): test labels.
x_pca (ndarray): reduced dataset by PCA.
x_ica (ndarray): reduced dataset by ICA.
x_kpca (ndarray): reduced dataset by KPCA.
x_rp (ndarray): reduced dataset by RP.
x_kmeans (ndarray): clusters produced by k-Means.
x_gmm (ndarray): clusters produced by Gaussian Mixture Models.
kwargs (dict): additional arguments to pass:
- layer1_nodes (int): number of neurons in first layer.
- layer2_nodes (int): number of neurons in second layer.
- learning_rate (float): learning rate.
Returns:
None.
"""
print('\n--------------------------')
print('NN')
print('--------------------------')
# Declare Neural Network and perform experiments on the original dataset
nn = NeuralNetwork(layer1_nodes=kwargs['layer1_nodes'],
layer2_nodes=kwargs['layer2_nodes'],
learning_rate=kwargs['learning_rate'])
nn.experiment(x_train, x_test, y_train, y_test)
print('\n--------------------------')
print('PCA + NN')
print('--------------------------')
# Declare Neural Network and perform experiments on the reduced dataset by PCA
nn = NeuralNetwork(layer1_nodes=kwargs['layer1_nodes'],
layer2_nodes=kwargs['layer2_nodes'],
learning_rate=kwargs['learning_rate'])
nn.experiment(x_pca[0], x_pca[1], y_train, y_test)
print('\n--------------------------')
print('ICA + NN')
print('--------------------------')
# Declare Neural Network and perform experiments on the reduced dataset by ICA
nn = NeuralNetwork(layer1_nodes=kwargs['layer1_nodes'],
layer2_nodes=kwargs['layer2_nodes'],
learning_rate=kwargs['learning_rate'])
nn.experiment(x_ica[0], x_ica[1], y_train, y_test)
print('\n--------------------------')
print('KPCA + NN')
print('--------------------------')
# Declare Neural Network and perform experiments on the reduced dataset by KPCA
nn = NeuralNetwork(layer1_nodes=kwargs['layer1_nodes'],
layer2_nodes=kwargs['layer2_nodes'],
learning_rate=kwargs['learning_rate'])
nn.experiment(x_kpca[0], x_kpca[1], y_train, y_test)
print('\n--------------------------')
print('RP+ NN')
print('--------------------------')
# Declare Neural Network and perform experiments on the reduced dataset by RP
nn = NeuralNetwork(layer1_nodes=kwargs['layer1_nodes'],
layer2_nodes=kwargs['layer2_nodes'],
learning_rate=kwargs['learning_rate'])
nn.experiment(x_rp[0], x_rp[1], y_train, y_test)
print('\n--------------------------')
print('KMEANS+ NN')
print('--------------------------')
# Declare Neural Network
nn = NeuralNetwork(layer1_nodes=kwargs['layer1_nodes'],
layer2_nodes=kwargs['layer2_nodes'],
learning_rate=kwargs['learning_rate'])
# Augment the original dataset by adding clusters produced by k-Means as features
x_kmeans_normalized = (x_kmeans[0] - np.mean(x_kmeans[0])) / np.std(
x_kmeans[0])
x_kmeans_normalized = np.expand_dims(x_kmeans_normalized, axis=1)
x_train_new = np.append(x_train, x_kmeans_normalized, axis=1)
x_kmeans_normalized = (x_kmeans[1] - np.mean(x_kmeans[1])) / np.std(
x_kmeans[1])
x_kmeans_normalized = np.expand_dims(x_kmeans_normalized, axis=1)
x_test_new = np.append(x_test, x_kmeans_normalized, axis=1)
# Perform experiments on it
nn.experiment(x_train_new, x_test_new, y_train, y_test)
print('\n--------------------------')
print('GMM+ NN')
print('--------------------------')
# Declare Neural Network
nn = NeuralNetwork(layer1_nodes=kwargs['layer1_nodes'],
layer2_nodes=kwargs['layer2_nodes'],
learning_rate=kwargs['learning_rate'])
# Augment the original dataset by adding clusters produced by Gaussian Mixture Models as features
x_gmm_normalized = (x_gmm[0] - np.mean(x_gmm[0])) / np.std(x_gmm[0])
x_gmm_normalized = np.expand_dims(x_gmm_normalized, axis=1)
x_train_new = np.append(x_train, x_gmm_normalized, axis=1)
x_gmm_normalized = (x_gmm[1] - np.mean(x_gmm[1])) / np.std(x_gmm[1])
x_gmm_normalized = np.expand_dims(x_gmm_normalized, axis=1)
x_test_new = np.append(x_test, x_gmm_normalized, axis=1)
# Perform experiments on it
nn.experiment(x_train_new, x_test_new, y_train, y_test)
def experiment(x_train, x_test, y_train, y_test):
"""Perform experiment.
Args:
x_train (ndarray): training data.
x_test (ndarray): test data.
y_train (ndarray): training labels.
y_test (ndarray): test labels.
Returns:
None.
"""
# Array of training sizes to plot the learning curves over.
training_sizes = np.arange(20, int(len(x_train) * 0.9), 10)
# K-Nearest Neighbor
print('\n--------------------------')
knn = KNN(k=1, weights='uniform', p=2)
knn.experiment(x_train,
x_test,
y_train,
y_test,
cv=10,
y_lim=0.3,
n_neighbors_range=np.arange(1, 50, 2),
p_range=np.arange(1, 20),
weight_functions=['uniform', 'distance'],
train_sizes=training_sizes)
# Support Vector Machines
print('\n--------------------------')
svm = SVM(c=1., kernel='rbf', degree=3, gamma=0.001, random_state=42)
svm.experiment(x_train,
x_test,
y_train,
y_test,
cv=10,
y_lim=0.2,
C_range=[1, 5] + list(range(10, 100, 20)) +
list(range(100, 1000, 50)),
kernels=['linear', 'poly', 'rbf'],
gamma_range=np.logspace(-7, 0, 50),
poly_degrees=[2, 3, 4],
train_sizes=training_sizes)
# Decision Trees
print('\n--------------------------')
dt = DecisionTree(max_depth=1, min_samples_leaf=1, random_state=42)
dt.experiment(x_train,
x_test,
y_train,
y_test,
cv=10,
y_lim=0.1,
max_depth_range=list(range(1, 50)),
min_samples_leaf_range=list(range(1, 30)),
train_sizes=training_sizes)
# AdaBoost
print('\n--------------------------')
boosted_dt = AdaBoost(n_estimators=50,
learning_rate=1.,
max_depth=3,
random_state=42)
boosted_dt.experiment(x_train,
x_test,
y_train,
y_test,
cv=10,
y_lim=0.2,
max_depth_range=list(range(1, 30)),
n_estimators_range=[1, 3, 5, 8] +
list(range(10, 100, 5)) + list(range(100, 1000, 50)),
learning_rate_range=np.logspace(-6, 1, 50),
train_sizes=training_sizes)
# Neural Networks
print('\n--------------------------')
nn = NeuralNetwork(alpha=0.01,
layer1_nodes=50,
layer2_nodes=30,
learning_rate=0.001,
max_iter=100)
nn.experiment(x_train,
x_test,
y_train,
y_test,
cv=10,
y_lim=0.1,
alpha_range=np.logspace(-5, 1, 30),
learning_rate_range=np.logspace(-4, 0, 50),
train_sizes=training_sizes)