def deepaugment_image_generator(X,
y,
policy,
batch_size=64,
augment_chance=0.5):
"""Yields batch of images after applying random augmentations from the policy
Each image is augmented by 50% chance. If augmented, one of the augment-chain in the policy is applied.
Which augment-chain to apply is chosen randomly.
Args:
X (numpy.array):
labels (numpy.array):
policy (list): list of dictionaries
Returns:
"""
if type(policy) == str:
policy_df = pd.read_csv(policy)
policy_df = policy_df[[
"aug1_type", "aug1_magnitude", "aug2_type", "aug2_magnitude",
"portion"
]]
policy = policy_df.to_dict(orient="records")
print("Policies are:")
print(policy)
print()
while True:
ix = np.arange(len(X))
np.random.shuffle(ix)
for i in range(len(X) // batch_size):
_ix = ix[i * batch_size:(i + 1) * batch_size]
_X = X[_ix]
_y = y[_ix]
tiny_batch_size = 4
aug_X = _X[0:tiny_batch_size]
aug_y = _y[0:tiny_batch_size]
for j in range(1, len(_X) // tiny_batch_size):
tiny_X = _X[j * tiny_batch_size:(j + 1) * tiny_batch_size]
tiny_y = _y[j * tiny_batch_size:(j + 1) * tiny_batch_size]
if np.random.rand() <= augment_chance:
aug_chain = np.random.choice(policy)
aug_chain[
"portion"] = 1.0 # last element is portion, which we want to be 1
hyperparams = list(aug_chain.values())
aug_data = augment_by_policy(tiny_X, tiny_y, *hyperparams)
aug_data["X_train"] = apply_default_transformations(
aug_data["X_train"])
aug_X = np.concatenate([aug_X, aug_data["X_train"]])
aug_y = np.concatenate([aug_y, aug_data["y_train"]])
else:
aug_X = np.concatenate([aug_X, tiny_X])
aug_y = np.concatenate([aug_y, tiny_y])
yield aug_X, aug_y
def getDataIteration(self):
if(self.trial_no==0):
print("Begin with no augmentation!!")
self.trial_hyperparams = ["rotate", 0.0, "rotate", 0.0,
"rotate", 0.0, "rotate", 0.0,
"rotate", 0.0, "rotate", 0.0,
"rotate", 0.0, "rotate", 0.0,
"rotate", 0.0, "rotate", 0.0]
else:
self.trial_hyperparams = self.controller.ask()
"""
self.trial_hyperparams = ['brighten',0.8921855570502091,'shear',0.857902058468807,
'invert',0.5446000610299714, 'brighten',0.9562927411137008,
'sharpen',0.8628617700194842,'emboss',0.12277382759592961,
'dropout',0.6623048596672579,'translate-x',0.9731839138940244,
'invert',0.7829575191193677,'vertical-flip', 0.9729114349444642]
"""
print("parameters:",self.trial_hyperparams)
augmented_data = augment_by_policy(
self.data["X_train"], self.data["y_train"], *self.trial_hyperparams
)
return augmented_data,self.trial_hyperparams
def evaluate(self, trial_no, trial_hyperparams):
"""Evaluates objective function
Trains the child model k times with same augmentation hyperparameters.
k is determined by the user by `opt_samples` argument.
Args:
trial_no (int): no of trial. needed for recording to notebook
trial_hyperparams (list)
Returns:
float: trial-cost = 1 - avg. rewards from samples
"""
augmented_data = augment_by_policy(
self.data["X_train"], self.data["y_train"], *trial_hyperparams
)
sample_rewards = []
for sample_no in range(1, self.opt_samples + 1):
self.child_model.load_pre_augment_weights()
# TRAIN
history = self.child_model.fit(self.data, augmented_data)
#
reward = self.calculate_reward(history)
sample_rewards.append(reward)
self.notebook.record(
trial_no, trial_hyperparams, sample_no, reward, history
)
trial_cost = 1 - np.mean(sample_rewards)
self.notebook.save()
log_and_print(
f"{str(trial_no)}, {str(trial_cost)}, {str(trial_hyperparams)}",
self.logging,
)
return trial_cost
def evaluate(self, trial_no, trial_hyperparams):
"""Evaluates objective function
Trains the child model k times with same augmentation hyperparameters.
k is determined by the user by `opt_samples` argument.
Args:
trial_no (int): no of trial. needed for recording to notebook
trial_hyperparams (list)
Returns:
float: trial-cost = 1 - avg. rewards from samples
"""
augmented_data = augment_by_policy(self.data["X_train"],
self.data["y_train"],
*trial_hyperparams)
sample_rewards = []
#pytorch
layers = 2
init_channels = 24
use_aux = True
epochs = 30
lr = 0.01
momentum = 0.995
weight_decay = 0.995
drop_path_prob = 0.2
genotype = "Genotype(normal=[[('dil_conv_3x3', 0), ('sep_conv_5x5', 1)], [('sep_conv_3x3', 1), ('avg_pool_3x3', 0)],[('dil_conv_3x3', 1), ('dil_conv_3x3', 0)], [('sep_conv_3x3', 3), ('skip_connect', 1)]], normal_concat=range(2, 6), reduce=[[('sep_conv_3x3', 1), ('dil_conv_5x5', 0)], [('skip_connect', 0), ('sep_conv_5x5', 1)], [('sep_conv_5x5', 1),('sep_conv_5x5', 0)], [('max_pool_3x3', 1), ('sep_conv_3x3', 0)]], reduce_concat=range(2, 6))"
model = AugmentCNN(self.input_size, self.input_channels, init_channels,
self.n_classes, layers, use_aux, genotype)
model = nn.DataParallel(model, device_ids='0').to(device)
# model size
mb_params = utils.param_size(model)
logger.info("Model size = {:.3f} MB".format(mb_params))
# weights optimizer
optimizer = torch.optim.SGD(model.parameters(),
lr,
momentum=momentum,
weight_decay=weight_decay)
a = 2 / 0
"""
for sample_no in range(1, self.opt_samples + 1):
self.child_model.load_pre_augment_weights()
# TRAIN
history = self.child_model.fit(self.data, augmented_data)
#
reward = self.calculate_reward(history)
sample_rewards.append(reward)
self.notebook.record(
trial_no, trial_hyperparams, sample_no, reward, history
)
"""
best_top1 = -9999
for epoch in range(epochs):
lr_scheduler.step()
drop_prob = drop_path_prob * epoch / epochs
model.module.drop_path_prob(drop_prob)
# training
train(train_loader, model, optimizer, criterion, epoch)
# validation
cur_step = (epoch + 1) * len(train_loader)
top1 = validate(valid_loader, model, criterion, epoch, cur_step)
# save
if best_top1 < top1:
best_top1 = top1
is_best = True
else:
is_best = False
print('best_top1:', best_top1)
#sample_rewards.append(reward)
#self.notebook.record(
# trial_no, trial_hyperparams, sample_no, reward, history
#)
#trial_cost = 1 - np.mean(sample_rewards)
#self.notebook.save()
log_and_print(
f"{str(trial_no)}, {str(trial_cost)}, {str(trial_hyperparams)}",
self.logging,
)
#return trial_cost
return best_top1