def get_valid_score(train_loader, valid_loader, n_epoch, step_size, weight_decay):
print("weight_decay: ", weight_decay)
net = KLayerNeuralNetwork(n_hidden_nodes=[50, 50], p_dropout=0.0, batch_norm=True, batch_norm_momentum=0.7)
scheduler = CyclicLR(eta_min=1e-5, eta_max=1e-1, step_size=step_size)
# =========================================================================
best_valid_acc = -np.inf
for epoch in range(n_epoch):
train(train_loader, net, weight_decay, scheduler)
valid_acc = evaluate(valid_loader, net)
if valid_acc > best_valid_acc:
best_valid_acc = valid_acc
print("[Evaluate] Valid. Acc.: %.3f%%" % (
valid_acc*100))
print("[Result] Valid. Acc.: %.3f%%" % (
best_valid_acc*100))
return best_valid_acc
train_loader = cifar10_DataLoader(train_data, batch_size=batch_size)
valid_loader = cifar10_DataLoader(valid_data, batch_size=batch_size)
test_loader = cifar10_DataLoader(test_data, batch_size=batch_size)
# ==================================================================
net = TwoLayerNeuralNetwork(n_hidden_nodes=[800], p_dropout=0.0)
# net = TwoLayerNeuralNetwork(n_hidden_nodes=[768])
ntrain = train_data['labels'].shape[0]
n_step_per_cycle = 3
ncycle = 3
n_epoch = ncycle * n_step_per_cycle * 2
iter_per_epoch = int(np.ceil(ntrain / batch_size))
step_size = n_step_per_cycle * iter_per_epoch
# weight_decay = 0.004454
# weight_decay = 0.005
weight_decay = 0.000
scheduler = CyclicLR(eta_min=1e-5, eta_max=1e-1, step_size=step_size)
# =========================================================================
print("--------- Train Schedule ---------")
print("ncycle: ", ncycle)
print("n_epoch: ", n_epoch)
print("step_size: ", step_size)
print("iter_per_epoch: ", iter_per_epoch)
print("n_step_per_cycle: ", n_step_per_cycle)
print("weight_decay: ", weight_decay)
print("--------- Train Schedule ---------")
best_valid_acc = -np.inf
for epoch in range(n_epoch):
train(train_loader, net, weight_decay, scheduler)
valid_acc = evaluate(valid_loader, net)
test_acc = evaluate(test_loader, net)
if valid_acc > best_valid_acc:
train_loader = cifar10_DataLoader(train_data, batch_size=batch_size)
valid_loader = cifar10_DataLoader(valid_data, batch_size=batch_size)
test_loader = cifar10_DataLoader(test_data, batch_size=batch_size)
# ==================================================================
net = TwoLayerNeuralNetwork(n_hidden_nodes=[50], p_dropout=0.0)
# net = TwoLayerNeuralNetwork(n_hidden_nodes=[768])
ntrain = train_data['labels'].shape[0]
n_step_per_cycle = 2
ncycle = 2
n_epoch = ncycle*n_step_per_cycle*2
iter_per_epoch = int(np.ceil(ntrain / batch_size))
step_size = n_step_per_cycle*iter_per_epoch
# weight_decay = 0.004454
# weight_decay = 0.005
weight_decay = 0.000
scheduler = CyclicLR(eta_min=1e-5, eta_max=1e-1, step_size=step_size)
# =========================================================================
print("--------- Train Schedule ---------")
print("ncycle: ", ncycle)
print("n_epoch: ", n_epoch)
print("step_size: ", step_size)
print("iter_per_epoch: ", iter_per_epoch)
print("n_step_per_cycle: ", n_step_per_cycle)
print("weight_decay: ", weight_decay)
print("--------- Train Schedule ---------")
best_valid_acc = -np.inf
jitr = ImageJitter()
for epoch in range(n_epoch):
loss = 0
total = 0
correct = 0