def task_dog_breed_identification(data_path, dataset):
images_path = os.path.join(data_path, dataset, 'images_all')
label_path = os.path.join(data_path, dataset, 'labels.csv')
test_path = os.path.join(data_path, dataset, 'test')
load_dataset = task.Dataset(images_path, label_file=label_path)
@ag.obj(learning_rate=ag.space.Real(0.3, 0.5),
momentum=ag.space.Real(0.90, 0.95),
wd=ag.space.Real(1e-6, 1e-4, log=True),
multi_precision=False)
class NAG(optim.NAG):
pass
classifier = task.fit(dataset=load_dataset,
net=ag.Categorical('standford_dog_resnext101_64x4d',
'standford_dog_resnet152_v1'),
optimizer=NAG(),
epochs=20,
final_fit_epochs=180,
num_trials=40,
ngpus_per_trial=8,
batch_size=48,
verbose=False,
ensemble=1)
test_dataset = task.Dataset(test_path, train=False, crop_ratio=0.65)
inds, probs, probs_all = classifier.predict(test_dataset,
set_prob_thresh=0.001)
ag.utils.generate_prob_csv(test_dataset,
probs_all,
custom='./submission.csv')
def test_tricks(combination):
dataset = 'shopee-iet'
data_dir = './'
download_shopee(data_dir, dataset)
target = config_choice(dataset, data_dir, combination)
classifier = task.fit(dataset=task.Dataset(target['dataset']),
net=target['net'],
optimizer=target['optimizer'],
epochs=target['epochs'],
ngpus_per_trial=target['ngpus_per_trial'],
num_trials=target['num_trials'],
batch_size=target['batch_size'],
verbose=True,
search_strategy='random',
tricks=target['tricks'],
lr_config=target['lr_config'],
plot_results=True)
test_dataset = task.Dataset(target['dataset'].replace(
'train', 'test/BabyPants'),
train=False,
scale_ratio_choice=[0.7, 0.8, 0.875])
inds, probs, probs_all = classifier.predict(test_dataset,
set_prob_thresh=0.001)
print(inds[0], probs[0], probs_all[0])
print('Top-1 val acc: %.3f' % classifier.results['best_reward'])
def test_ensemble():
dataset = task.Dataset(name='FashionMNIST')
test_dataset = task.Dataset(name='FashionMNIST', train=False)
classifier = task.fit(dataset,
epochs=1,
ngpus_per_trial=1,
verbose=False,
ensemble=2)
test_acc = classifier.evaluate(test_dataset)
def test_classifier_save_load():
dataset = task.Dataset(name='FashionMNIST')
test_dataset = task.Dataset(name='FashionMNIST', train=False)
classifier = task.fit(dataset,
epochs=1,
ngpus_per_trial=1,
verbose=False)
classifier.save('classifier.ag')
classifier2 = task.Classifier.load('classifier.ag')
test_acc = classifier2.evaluate(test_dataset)
def main():
opt = parse_args()
if not os.path.exists(opt.dataset):
os.mkdir(opt.dataset)
dataset_path = os.path.join(opt.data_dir, opt.dataset)
local_path = os.path.dirname(__file__)
output_directory = os.path.join(opt.dataset, 'checkpoint/')
filehandler = logging.FileHandler(os.path.join(opt.dataset, 'summary.log'))
streamhandler = logging.StreamHandler()
logger = logging.getLogger('')
logger.setLevel(logging.INFO)
logger.addHandler(filehandler)
logger.addHandler(streamhandler)
logging.info(opt)
target = config_choice(opt.data_dir, opt.dataset)
load_dataset = task.Dataset(target['dataset'])
classifier = task.fit(dataset=load_dataset,
output_directory=output_directory,
net=target['net'],
optimizer=target['optimizer'],
tricks=target['tricks'],
lr_config=target['lr_config'],
resume=opt.resume,
epochs=opt.num_epochs,
ngpus_per_trial=opt.ngpus_per_trial,
num_trials=opt.num_trials,
batch_size=opt.batch_size,
verbose=True,
plot_results=True)
summary = classifier.fit_summary(output_directory=opt.dataset, verbosity=4)
logging.info('Top-1 val acc: %.3f' % classifier.results['best_reward'])
logger.info(summary)
if opt.submission:
test_dataset = task.Dataset(os.path.join(opt.data_dir, opt.dataset, 'test'), train=False)
inds, probs, probs_all, value = predict_details(test_dataset, classifier, load_dataset)
generate_csv_submission(dataset_path, opt.dataset, local_path, inds, probs_all, value, opt.custom)
import math
import autogluon.core as ag
from autogluon.mxnet.optimizer import SGD
from autogluon.extra.model_zoo import EfficientNet
from autogluon.vision import ImageClassification as task
@ag.obj(
width_coefficient=ag.Categorical(1.1, 1.2),
depth_coefficient=ag.Categorical(1.1, 1.2),
)
class EfficientNetB1(EfficientNet):
def __init__(self, width_coefficient, depth_coefficient):
input_factor = math.sqrt(2.0 / (width_coefficient**2) /
depth_coefficient)
input_size = math.ceil((224 * input_factor) / 32) * 32
super().__init__(width_coefficient=width_coefficient,
depth_coefficient=depth_coefficient,
input_size=input_size)
results = task.fit('imagenet',
net=EfficientNetB1(),
search_strategy='grid',
optimizer=SGD(learning_rate=1e-1, momentum=0.9, wd=1e-4),
batch_size=32)
print(results)
if len(network_list) > 1:
nets = ag.Categorical(*network_list)
else:
nets = network_list[0]
classifier = task.fit(dataset,
net=nets,
optimizer=optimizer,
search_strategy='skopt',
search_options={
'base_estimator': 'GP',
'acq_func': 'EI'
},
epochs=args.epochs,
batch_size=args.batch_size,
metric='accuracy',
lr_config=lr_config,
verborse=True,
plot_results=True,
visualizer='tensorboard',
ngpus_per_trial=args.ngpus_per_trial,
num_trials=args.num_trials,
final_fit_epochs=args.final_fit_epochs,
use_pretrained=True,
batch_norm=False,
output_directory=os.environ["SM_OUTPUT_DATA_DIR"])
test_dataset = create_dataset(os.environ["SM_CHANNEL_TEST"], train=False)
test_acc = classifier.evaluate(test_dataset)
logging.info('Test accuracy: %f' % test_acc)