count = 0
# Shuffle training data
np.random.shuffle(training_data)
# Define structure of optimal network
HL = 2
HN1, HN2 = 20, 4
EPOCHS = 100
BATCH_SIZE = 50
LR = 0.0007
# Instantiate the network and prepare data
avg_mse = 1
while avg_mse > 0.9:
net = Net(HN1, HN2)
training_inputs = training_data[:, 0:4]
training_labels = training_data[:, 4:]
test_inputs = testing_data[:, 0:4]
test_labels = testing_data[:, 4:]
# Train and test the network
train(net, training_inputs, training_labels, EPOCHS, LR, BATCH_SIZE)
avg_mse, predictions_online, predictions_offline = test(
test_inputs, test_labels, net)
print(avg_mse)
predictions_online_inverse_transform = scaler_test.inverse_transform(
predictions_online)
predictions_offline_inverse_transform = scaler_test.inverse_transform(
predictions_offline)
# k-fold cross validation training loop
HL = 2
HN1 = 10
HN2 = 10
EPOCHS = 50
BATCH_SIZE = 50
LR = [
0.0001, 0.0002, 0.0003, 0.0004, 0.0005, 0.0006, 0.0007, 0.0008, 0.0009,
0.001, 0.002, 0.003, 0.004, 0.005, 0.006, 0.007, 0.008, 0.009, 0.01, 0.02,
0.03, 0.04, 0.05, 0.06, 0.07, 0.08, 0.09, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6,
0.7, 0.8, 0.9, 1
]
MODELS = {}
net = Net(HN1, HN2)
init_state = copy.deepcopy(net.state_dict())
for lr in LR:
MSEs = []
for index, subset in enumerate(subset_train_list):
subset.value = np.array(subset.value)
subset_test_list[index].value = np.array(subset_test_list[index].value)
net.load_state_dict(init_state)
training_inputs = subset.value[:, 0:5]
training_labels = subset.value[:, 5:]
test_inputs = subset_test_list[index].value[:, 0:5]
test_labels = subset_test_list[index].value[:, 5:]
train(net, training_inputs, training_labels, EPOCHS, lr, BATCH_SIZE)
avg_mse = test(test_inputs, test_labels, net)
HL = 2
HN = [(2, 2), (2, 4), (2, 8), (2, 12),
(2, 16), (2, 20), (4, 2), (4, 4), (4, 8), (4, 12), (4, 16), (4, 20),
(8, 2), (8, 4), (8, 8), (8, 12), (8, 16), (8, 20), (12, 2), (12, 4),
(12, 8), (12, 12), (12, 16), (12, 20), (16, 2), (16, 4), (16, 8),
(16, 12), (16, 16), (16, 20), (20, 2), (20, 4), (20, 8), (20, 12),
(20, 16), (20, 20)]
EPOCHS = 500
BATCH_SIZE = 50
LR = 0.001
MODELS = {}
print(training_data)
for h1, h2 in HN:
net = Net(h1, h2)
init_state = copy.deepcopy(net.state_dict())
net.load_state_dict(init_state)
training_inputs = training_data[:, 0:4]
training_labels = training_data[:, 4:]
test_inputs = testing_data[:, 0:4]
test_labels = testing_data[:, 4:]
E_opt, opt_epochs = train(net, training_inputs, training_labels,
test_inputs, test_labels, EPOCHS, LR, BATCH_SIZE)
MODELS['{b}_{x}-{y}_{z}'.format(b=HL, x=h1, y=h2, z=opt_epochs)] = E_opt
with open(
'Data3/Search/manual_search_results_{x}HL_hn-e_50rep.csv'.format(x=HL),
'w') as f:
count = 0
# Shuffle training data
np.random.shuffle(training_data)
# Define structure of optimal network
HL = 2
HN1, HN2 = 4, 8
EPOCHS = 152
BATCH_SIZE = 300
LR = 0.006
# Instantiate the network and prepare data
avg_mse = 1
while avg_mse > 0.09:
net = Net(HN1, HN2)
training_inputs = training_data[:, 0:5]
training_labels = training_data[:, 5:]
test_inputs = testing_data[:, 0:5]
test_labels = testing_data[:, 5:]
# Train and test the network
train(net, training_inputs, training_labels, EPOCHS, LR, BATCH_SIZE)
avg_mse, predictions_online, predictions_offline = test(
test_inputs, test_labels, net)
print(avg_mse)
predictions_online_inverse_transform = scaler_test.inverse_transform(
predictions_online)
predictions_offline_inverse_transform = scaler_test.inverse_transform(
predictions_offline)