def weight_decay_test():
epoch_list = [x for x in range(1000, 15001) if x % 1000 == 0]
# test_errors = {}
train_errors = {}
test_rate = []
for epochs in epoch_list:
train_err = []
cnn = CifarLeNet(lr=0.001, epochs=epochs, batch_size=128, train_data=cifar_train,
test_data=cifar_test, wd=0.004, decay_lr=False,
decay_w=True, optimizer='adam', seed=100)
cnn.train(train_err)
train_errors[str(epochs)] = train_err
cnn.test_eval(epochs, test_rate)
cnn.reset()
with open('pickled_data/test_acc_epochs_wd', 'wb') as f:
pickle.dump([train_errors, test_rate], f)
return test_rate
def batch_sizes():
sizes = [10, 25, 50, 64, 100, 200, 300, 400, 500, 750, 1000, 2000]
# test_errors = {}
train_errors = {}
test_rate = []
for bs in sizes:
train_err = []
cnn = CifarLeNet(lr=0.001, epochs=10000, batch_size=bs, test_data=cifar_test,
train_data=cifar_train, wd=0.004, decay_lr=False,
decay_w=True, optimizer='adam', seed=100)
cnn.train(train_err)
train_errors[str(bs)] = train_err
cnn.test_eval(bs, test_rate)
cnn.reset()
with open('pickled_data/test_acc_bs', 'wb') as f:
pickle.dump([train_errors, test_rate], f)
return test_rate
def lr_test():
lr_rates = [0.001, 0.01, 0.1, 0.5, 1.0]
# test_errors = {}
train_errors = {}
test_rate = []
for rate in lr_rates:
train_err = []
cnn = CifarLeNet(lr=rate, epochs=10000, batch_size=128, train_data=cifar_train,
test_data=cifar_test, wd=0.004, decay_lr=False,
decay_w=True, optimizer='adam', seed=100)
cnn.train(train_err)
train_errors[str(rate)] = train_err
cnn.test_eval(rate, test_rate)
cnn.reset()
with open('pickled_data/test_acc_lrs', 'wb') as f:
pickle.dump([train_errors, test_rate], f)
return test_rate
def lrs(rates, filename):
train_errors = {}
for lr_rate in rates:
acc_list = []
cnn = CifarLeNet(lr=lr_rate,
epochs=5000,
batch_size=500,
train_data=cifar_train,
test_data=cifar_test,
wd=0.004,
decay_lr=False,
decay_w=False,
optimizer='adam',
seed=100)
cnn.train(acc_list)
cnn.reset()
train_errors[str(lr_rate)] = acc_list
with open('pickled_data/' + filename, 'wb') as f:
pickle.dump(train_errors, f)
def decay_lr(filename, start_rate=0.01):
train_errors = {}
print('Decay')
acc_list = []
cnn = CifarLeNet(lr=start_rate,
epochs=10000,
batch_size=128,
train_data=cifar_train,
test_data=cifar_test,
wd=0.004,
decay_lr=True,
decay_w=False,
optimizer='adam',
seed=100)
cnn.train(acc_list)
cnn.reset()
train_errors['decay'] = acc_list
with open('pickled_data/' + filename, 'wb') as f:
pickle.dump(train_errors, f)
return train_errors
def algos(optimizers, filename, m=False, m_val=0.5):
if not m:
train_errors = {}
for algo in optimizers:
acc_list = []
cnn = CifarLeNet(lr=0.0001,
epochs=5000,
batch_size=500,
train_data=cifar_train,
test_data=cifar_test,
wd=0.004,
decay_lr=False,
decay_w=False,
optimizer=algo,
seed=100)
cnn.train(acc_list)
cnn.reset()
train_errors[algo] = acc_list
with open('pickled_data/' + filename, 'wb') as f:
pickle.dump(train_errors, f)
else:
train_errors = {}
for value in m_val:
acc_list = []
cnn = CifarLeNet(lr=0.0001,
epochs=5000,
batch_size=500,
train_data=cifar_train,
test_data=cifar_test,
wd=0.004,
decay_lr=False,
decay_w=False,
optimizer='momentum',
seed=100,
momentum=value)
cnn.train(acc_list)
cnn.reset()
train_errors['momentum' + str(value)] = acc_list
with open('pickled_data/' + filename, 'wb') as f:
pickle.dump(train_errors, f)