def test_nn(hid_node, learning_rate, activation_function, pruning_choice):
acc_stat = []
con_mat_stat = []
for i in range(len(fold_groups)):
print("Round: " + str(i + 1))
# Select test set and train data
test_group_idx = i
test_X, test_y = fold_groups[i]
train_group_idx = list(five_sets - {i})
train_X = [fold_groups[idx][0] for idx in train_group_idx]
train_y = [fold_groups[idx][1] for idx in train_group_idx]
train_X = np.concatenate(train_X)
train_y = np.concatenate(train_y)
# Initialize model
in_node = train_X.shape[1]
out_node = 1
NN = NeuralNetwork(in_node,
hid_node,
out_node,
activation=activation_function,
seed=None)
# Train the model
if pruning_choice:
NN.fit_model(train_X,
train_y,
learning_epochs=20,
learning_rate=learning_rate,
pruning=pruning_choice,
pruning_threshold=0.2,
pruning_norm='L1',
pruning_epoch=10)
else:
NN.fit_model(train_X,
train_y,
learning_epochs=20,
learning_rate=learning_rate,
pruning=pruning_choice)
acc = NN.accuracy(test_X, test_y)
# Compute accuracy and confusion matrix
acc = 0
y_pred = NN.predict_result(test_X)
con_mat = np.zeros((2, 2))
for i in range(len(y_pred)):
con_mat[int(y_pred[i]), int(test_y[i])] += 1
if test_y[i] == y_pred[i]:
acc += 1
con_mat_stat.append(con_mat)
acc = acc / len(y_pred)
acc_stat.append(acc)
print('\nACCURACY: ', acc)
print('CONFUSION MATRIX: \n', con_mat)
print("\n")
print("\nACCURACY AVERAGE: ", np.average(acc_stat))
print("CONFUSION MATRIX AVERAGE: \n", np.average(con_mat_stat, axis=0))
print("Performing Principal Component Analysis...")
# Perform Principal Component Analysis on the train and test data
for j in range(len(train_data)):
train_data[j] = pca.fit_transform(train_data[j]).ravel()
for j in range(len(test_data)):
test_data[j] = pca.fit_transform(test_data[j]).ravel()
accuracy = 0
# create an instance of the network
nn = NeuralNetwork(1000, 500, 3, 0.01)
print("Training...")
# iterate N times perform training at each step
for epoch in range(100):
for j in range(len(train_data)):
nn.train(train_data[j], labels[j])
accuracy = nn.accuracy(test_data, test_labels)
print("\tepoch {i}, accuracy = {accuracy}".format(i=epoch + 1,
accuracy=accuracy))
# ask the user if they'd like to save the trained model, then the appropriate action is taken
save = input("Save current model with accuracy %.4f? [Y/N]: " % accuracy)
if save == "Y":
model_file_path = "models/model.json"
nn.write_to_file(model_file_path)
print("Model saved to %s" % model_file_path)
else:
print("Model not saved.")
print("Done")
hid_node = 7
out_node = 3
NN = NeuralNetwork(in_node,
hid_node,
out_node,
activation='sigmoid',
seed=None)
# Train the model
NN.fit_model(train_X,
train_y,
learning_epochs=500,
learning_rate=0.2,
pruning=False,
verbose=False)
acc = NN.accuracy(test_X, test_y)
# compute accuracy and confusion matrix
acc = 0
y_pred = NN.predict_result(test_X)
con_mat = np.zeros((3, 3))
for i in range(len(y_pred)):
con_mat[int(y_pred[i]), int(test_y[i])] += 1
if test_y[i] == y_pred[i]:
acc += 1
con_mat_stat.append(con_mat)
acc = acc / len(y_pred)
acc_stat.append(acc)
print('\nACCURACY: ', acc)
print("Performing Principal Component Analysis...")
# Perform Principal Component Analysis on the train and test data
for j in range(len(train_data)):
train_data[j] = pca.fit_transform(train_data[j]).ravel()
for j in range(len(test_data)):
test_data[j] = pca.fit_transform(test_data[j]).ravel()
accuracy = 0
# create an instance of the network
nn = NeuralNetwork(1000, 500, 3, 0.01)
print("Training...")
# iterate N times perform training at each step
for epoch in range(100):
for j in range(len(train_data)):
nn.train(train_data[j], labels[j])
accuracy = nn.accuracy(train_data, labels)
print("\tepoch {i}, accuracy = {accuracy}".format(i=epoch + 1,
accuracy=accuracy))
# ask the user if they'd like to save the trained model, then the appropriate action is taken
save = input("Save current model with accuracy %.4f? [Y/N]: " % accuracy)
if save == "Y":
model_file_path = "models/model.json"
nn.write_to_file(model_file_path)
print("Model saved to %s" % model_file_path)
else:
print("Model not saved.")
print("Done")
