def verify_on_full_data():
model = ThreeLayerConvNet(num_classes=2, weight_scale=0.001, hidden_dim=500, reg=0.01, dropout=0.4)
data = get_CIFAR2_data()
solver = Solver(model, data,
num_epochs=10, batch_size=50,
update_rule='adam',
optim_config={
'learning_rate': 1e-3,
},
verbose=True, print_every=20)
solver.train()
def verify_model_on_small_data():
data = get_CIFAR2_data()
for k, v in data.items():
print('% s: ' % k, v.shape)
np.random.seed(231)
num_train = 100
small_data = {
'X_train': data['X_train'][:num_train],
'y_train': data['y_train'][:num_train],
'X_val': data['X_val'],
'y_val': data['y_val'],
}
model = cnn.ThreeLayerConvNet(weight_scale=1e-2)
solver = Solver(model,
small_data,
num_epochs=15,
batch_size=50,
update_rule='adam',
optim_config={
'learning_rate': 1e-3,
},
verbose=True,
print_every=1)
solver.train()
plt.subplot(2, 1, 1)
plt.plot(solver.loss_history, 'o')
plt.xlabel('iteration')
plt.ylabel('loss')
plt.subplot(2, 1, 2)
plt.plot(solver.train_acc_history, '-o')
plt.plot(solver.val_acc_history, '-o')
plt.legend(['train', 'val'], loc='upper left')
plt.xlabel('epoch')
plt.ylabel('accuracy')
plt.show()
def verify_on_full_data_without_dropout():
model = ThreeLayerConvNet(num_classes=2,
weight_scale=0.001,
hidden_dim=500,
reg=0,
dropout=0)
data = get_CIFAR2_data()
solver = Solver(model,
data,
num_epochs=10,
batch_size=50,
update_rule='adam',
optim_config={
'learning_rate': 1e-3,
},
verbose=True,
print_every=20)
solver.train()
plt.subplot(2, 1, 1)
plt.plot(solver.loss_history, 'o')
plt.xlabel('iteration')
plt.ylabel('loss')
plt.subplot(2, 1, 2)
plt.plot(solver.train_acc_history, '-o')
plt.plot(solver.val_acc_history, '-o')
plt.legend(['train', 'val'], loc='upper left')
plt.xlabel('epoch')
plt.ylabel('accuracy')
plt.show()
grid = visualize_grid(model.params['W1'].transpose(0, 2, 3, 1))
plt.imshow(grid.astype('uint8'))
plt.axis('off')
plt.gcf().set_size_inches(5, 5)
plt.show()
from code_base.solver import Solver
data = get_CIFAR2_data()
for k, v in data.items():
print('%s: ' % k, v.shape)
model = ThreeLayerConvNet(num_classes=2,
weight_scale=0.001,
hidden_dim=500,
reg=0.001)
solver = Solver(model,
data,
num_epochs=1,
batch_size=50,
update_rule='adam',
optim_config={
'learning_rate': 1e-3,
},
verbose=True,
print_every=20)
print('start train')
solver.train()
print('finish train')
plt.subplot(2, 1, 1)
plt.plot(solver.loss_history, 'o')
plt.xlabel('iteration')
plt.ylabel('loss')
plt.subplot(2, 1, 2)
plt.plot(solver.train_acc_history, '-o')
# A bit of setup
from __future__ import print_function
from code_base.data_utils import *
# Load the (preprocessed) CIFAR2 (airplane and bird) data.
from code_base.solver import Solver
data = get_CIFAR2_data()
for k, v in data.items():
print('%s: ' % k, v.shape)
model = load_model('./models/cnn_model_conv64_fil5_fc512_numc2_dropout_0.5.p')
model.use_dropout = False
print(
'Test Accuracy: ',
Solver(model, data=data).check_accuracy(data['X_val'],
data['y_val'],
batch_size=500))
'X_val': data['X_val'],
'y_val': data['y_val'],
}
to_load = input('\nDo you want to load pre-trained model? [y/n]: ')
if to_load == 'y' or to_load == 'Y':
model = load_model('./models/cnn_model_conv64_fil5_fc512_numc2_dropout_0.5.p')
else:
model = ThreeLayerConvNet(num_classes=2, weight_scale=0.001, hidden_dim=512, reg=0.000001, num_filters=64,
filter_size=5, dropout=0.5)
solver = Solver(model, data,
num_epochs=1, batch_size=128,
update_rule='adam',
optim_config={
'learning_rate': 1e-4,
},
lr_decay=0.8,
num_train_samples=1000,
verbose=True, print_every=1)
start = datetime.datetime.now()
solver.train()
end = datetime.datetime.now()
print('Total time taken: ', end - start)
to_save = input('\nDo you want to save this? [y/n]: ')
if to_save == 'y' or to_save == 'Y':
save_model(model, './models/cnn_model_conv64_fil5_fc512_numc2_dropout_0.5.p')
plt.subplot(2, 1, 1)
plt.plot(solver.loss_history, 'o')