cs.add_conditions(aug_space.get_conditions())
for estimator_id in algorithm_candidates:
sub_cs = get_model_config_space(estimator_id,
include_estimator=False,
include_aug=False)
parent_hyperparameter = {
'parent': estimator_choice,
'value': estimator_id
}
cs.add_configuration_space(estimator_id,
sub_cs,
parent_hyperparameter=parent_hyperparameter)
return cs
cs = get_pipeline_config_space(['resnet34', 'mobilenet'])
dataset = 'cifar10'
data_dir = 'data/img_datasets/%s/' % dataset
image_data = ImageDataset(data_path=data_dir, train_val_split=True)
hpo_evaluator = DLEvaluator(cs.get_default_configuration(),
IMG_CLS,
scorer=get_metric('acc'),
dataset=image_data,
device='cuda',
image_size=32,
seed=1)
hpo_evaluator(cs.get_default_configuration())
transforms.RandomResizedCrop(560),
transforms.RandomCrop(331),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
'val':
transforms.Compose([
transforms.Resize(560),
transforms.CenterCrop(331),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
}
data_dir = 'data/img_datasets/hymenoptera_data'
image_data = ImageDataset(data_path=data_dir, data_transforms=data_transforms)
print(image_data.train_dataset.classes)
print(image_data.val_dataset.classes)
image_data.load_test_data()
print(image_data.test_dataset.classes)
image_data = ImageDataset(data_path=data_dir,
data_transforms=data_transforms,
train_val_split=True,
val_split_size=0.3)
print(image_data.train_dataset.classes)
print(image_data.train_sampler, image_data.val_sampler)
import sys
import argparse
sys.path.append(os.getcwd())
from mindware.datasets.image_dataset import ImageDataset
from mindware.estimators import ImageClassifier
parser = argparse.ArgumentParser()
parser.add_argument('--n_jobs', type=int, default=3)
parser.add_argument('--dataset', type=str, default='extremely_small')
args = parser.parse_args()
n_jobs = args.n_jobs
dataset = args.dataset
print('n_jobs is set to %d.' % n_jobs)
data_dir = 'data/img_datasets/%s/' % dataset
image_data = ImageDataset(data_path=data_dir, train_val_split=True)
clf = ImageClassifier(
time_limit=10000000,
include_algorithms=['mobilenet', 'resnet34', 'efficientnet'],
evaluation='partial',
image_size=32,
ensemble_method='ensemble_selection',
skip_profile=True,
n_jobs=n_jobs)
clf.fit(image_data)
image_data.set_test_path(data_dir)
print(clf.predict_proba(image_data))
pred = clf.predict(image_data)
from mindware.components.models.img_classification.resnext import ResNeXtClassifier
from mindware.components.models.img_classification.senet import SENetClassifier
from mindware.components.models.img_classification.nasnet import NASNetClassifier
from mindware.components.models.img_classification.mobilenet import MobileNettClassifier
from mindware.components.models.img_classification.efficientnet import EfficientNetClassifier
from mindware.components.models.img_classification.resnet110_32 import ResNet110_32Classifier
from mindware.components.models.img_classification.densenet190_32 import DenseNet190_32Classifier
from mindware.components.models.img_classification.densenet100_32 import DenseNet100_32Classifier
phase = 'test'
if phase == 'fit':
# data_dir = 'data/img_datasets/hymenoptera_data/'
data_dir = 'data/img_datasets/cifar10/'
# data_dir = 'data/img_datasets/dogs-vs-cats/'
image_data = ImageDataset(data_path=data_dir, train_val_split=True, image_size=32)
clf = ImageClassifier(time_limit=3600 * 10,
include_algorithms=['resnet44_32'],
evaluation='partial',
max_epoch=120,
skip_profile=True,
ensemble_method='ensemble_selection',
n_jobs=3)
clf.fit(image_data, opt_method='whatever')
image_data.set_test_path(data_dir)
print(clf.score(image_data, mode='val'))
print(clf.score(image_data))
pred = clf.predict(image_data)
timestamp = time.time()
with open('es_output_%s.pkl' % timestamp, 'wb') as f:
pkl.dump(pred, f)
import os
import sys
from torchvision import transforms
from sklearn.metrics import accuracy_score
sys.path.append(os.getcwd())
from mindware.datasets.image_dataset import ImageDataset
from mindware.estimators import ImageClassifier
from mindware.components.models.img_classification import add_classifier
from mindware.components.models.img_classification.resnext import ResNeXtClassifier
# Add user-defined classifier
add_classifier(ResNeXtClassifier)
data_dir = 'data/img_datasets/hymenoptera_data/'
image_data = ImageDataset(data_path=data_dir)
clf = ImageClassifier(time_limit=7200,
include_algorithms=['ResNeXtClassifier'],
ensemble_method='ensemble_selection',
config_file_path='tiny_cs.txt')
clf.fit(image_data)
image_data.set_test_path(data_dir)
print(clf.predict_proba(image_data))
print(clf.predict(image_data))
'resnet50', 'resnet101', 'densenet121']
parser.add_argument('--networks', type=str, default=','.join(networks_template))
args = parser.parse_args()
n_jobs = args.n_jobs
dataset = args.dataset
networks = args.networks.split(',')
time_limit = args.time_limit
opt_method = args.opt_method
evaluation = args.eval
max_epoch = args.max_epoch
img_size = args.img_size
print('n_jobs is set to %d.' % n_jobs)
print('networks included', networks)
data_dir = 'data/img_datasets/%s/' % dataset
image_data = ImageDataset(data_path=data_dir, train_val_split=True, image_size=img_size, val_split_size=0.1)
clf = ImageClassifier(time_limit=time_limit,
include_algorithms=networks,
evaluation=evaluation,
ensemble_method=None,
skip_profile=True,
max_epoch=max_epoch,
n_jobs=n_jobs)
image_data.set_test_path(data_dir)
if opt_method == 'ours':
clf.fit(image_data)
else:
clf.fit(image_data, opt_method=opt_method)
res = clf.get_runtime_history()