def _init_conf(self, is_search_or_eval:bool, clean_expdir:bool)->Config:
config_filename = self.config_filename
conf = common.common_init(config_filepath=config_filename,
param_args=['--common.experiment_name', self.get_expname(is_search_or_eval),
], clean_expdir=clean_expdir)
return conf
def test_petridish_zero_model():
conf = common_init(config_filepath='confs/petridish_cifar.yaml')
conf_search = conf['nas']['search']
model_desc = conf_search['model_desc']
model_desc_builder = ModelDescBuilder()
model_desc = model_desc_builder.build(model_desc)
m = Model(model_desc, False, True)
y, aux = m(torch.rand((1, 3, 32, 32)))
assert isinstance(y, torch.Tensor) and y.shape == (1, 10) and aux is None
def test_darts_zero_model():
conf = common_init(config_filepath='confs/darts_cifar.yaml')
conf_search = conf['nas']['search']
model_desc = conf_search['model_desc']
macro_builder = MacroBuilder(model_desc, aux_tower=False)
model_desc = macro_builder.build()
m = Model(model_desc, False, True)
y, aux = m(torch.rand((1, 3, 32, 32)))
assert isinstance(y, torch.Tensor) and y.shape==(1,10) and aux is None
def _init(self, suffix: str) -> Config:
config_filename = self.config_filename
if self.toy_config_filename:
config_filename += ';' + self.toy_config_filename
conf = common_init(config_filepath=config_filename,
param_args=[
'--common.experiment_name',
self.base_name + f'_{suffix}',
])
return conf
def main():
conf = common_init(config_filepath='confs/algos/resnet.yaml')
provider = Cifar10Provider(conf['dataset'])
train_transform, test_transform = provider.get_transforms()
trainset, testset = provider.get_datasets(load_train=True,
load_test=True,
transform_train=None,
transform_test=None)
trainset = MetaDataset(trainset, transform=train_transform)
for x, y, meta in trainset:
print(x, y, meta)
def main():
#6, 7, 9, 10, 16
#model = model_builder.build(model_builder.EXAMPLE_DESC_MATRIX, model_builder.EXAMPLE_VERTEX_OPS)
nsds = Nasbench101Dataset('~/dataroot/nasbench_ds/nasbench_full.pkl')
conf = common_init(config_filepath='confs/algos/nasbench101.yaml')
conf_eval = conf['nas']['eval']
conf_loader = conf_eval['loader']
conf_trainer = conf_eval['trainer']
model = nsds.create_model(5) # 401277 is same model as example
data_loaders = data.get_data(conf_loader)
trainer = Trainer(conf_trainer, model)
trainer.fit(data_loaders)
from archai.nas.arch_trainer import ArchTrainer
from archai.common.common import common_init
from archai.nas import search
from archai.petridish.petridish_micro_builder import PetridishMicroBuilder
if __name__ == '__main__':
conf = common_init(
config_filepath='confs/petridish_cifar.yaml',
param_args=['--common.experiment_name', 'petridish_cifar_search'])
# region config
conf_search = conf['nas']['search']
# endregion
micro_builder = PetridishMicroBuilder()
trainer_class = ArchTrainer
search.search_arch(conf_search, micro_builder, trainer_class)
exit(0)
def main():
# accept search and eval scripts to run
# config file can be supplied using --config
parser = argparse.ArgumentParser(description='NAS E2E Runs')
parser.add_argument('--search-script',
type=str,
default='scripts/darts/cifar_search.py',
help='Search script to run')
parser.add_argument('--eval-script',
type=str,
default='scripts/darts/cifar_eval.py',
help='Eval script to run')
parser.add_argument('--exp_prefix',
type=str,
default='darts',
help='Experiment prefix to use')
args, extra_args = parser.parse_known_args()
# load config to some of the settings like logdir
conf = common_init(use_args=True)
logdir = get_conf_common()['logdir']
assert logdir
# get script, resume flag and experiment dir for search
search_script = args.search_script
resume = conf['nas']['search']['resume']
search_script = utils.full_path(search_script.strip())
experiment_name = args.exp_prefix + '_' + Path(search_script).stem
experiment_dir = os.path.join(logdir, experiment_name)
# see if search has already produced the output
final_desc_filepath = os.path.join(
experiment_dir, conf['nas']['search']['final_desc_filename'])
if not resume or not os.path.exists(final_desc_filepath):
print(f'Starting {search_script}...')
result = subprocess.run([
'python', search_script, '--config', conf.config_filepath,
'--config-defaults', conf.config_defaults_filepath,
'--common.experiment_name', experiment_name
])
print(f'Script {search_script} returned {result.returncode}')
if result.returncode != 0:
exit(result.returncode)
else:
print(
f'Search is skipped because file {final_desc_filepath} already exists'
)
# get script, resume flag and experiment dir for eval
eval_script = args.eval_script
resume = conf['nas']['eval']['resume']
eval_script = utils.full_path(eval_script.strip())
experiment_name = args.exp_prefix + '_' + Path(eval_script).stem
experiment_dir = os.path.join(logdir, experiment_name)
# if eval has already produced the output, skip eval run
model_filepath = os.path.join(experiment_dir,
conf['nas']['eval']['save_filename'])
if not resume or not os.path.exists(model_filepath):
# copy output of search to eval folder
# TODO: take final_desc_filename from eval config
os.makedirs(experiment_dir, exist_ok=True)
shutil.copy2(final_desc_filepath, experiment_dir)
print(f'Starting {eval_script}...')
result = subprocess.run([
'python', eval_script, '--config', conf.config_filepath,
'--config-defaults', conf.config_defaults_filepath,
'--common.experiment_name', experiment_name
])
print(f'Script {eval_script} returned {result.returncode}')
if result.returncode != 0:
exit(result.returncode)
else:
print(f'Eval is skipped because file {model_filepath} already exists')
print('Search and eval done.')
exit(0)
from archai.common.common import common_init
from archai.data_aug.search import search
if __name__ == '__main__':
conf = common_init(config_filepath='confs/wresnet40x2_cifar10_b512.yaml')
search(conf)
# Copyright (c) Microsoft Corporation.
# Licensed under the MIT license.
from archai.nas.model_desc import ModelDesc
from archai.common.common import common_init
from archai.nas.model import Model
from archai.algos.petridish.petridish_model_desc_builder import PetridishModelBuilder
from archai.common.model_summary import summary
conf = common_init(config_filepath='confs/petridish_cifar.yaml',
param_args=['--common.experiment_name', 'petridish_run2_seed42_eval'])
conf_eval = conf['nas']['eval']
conf_model_desc = conf_eval['model_desc']
conf_model_desc['n_cells'] = 14
template_model_desc = ModelDesc.load('$expdir/final_model_desc.yaml')
model_builder = PetridishModelBuilder()
model_desc = model_builder.build(conf_model_desc, template=template_model_desc)
mb = PetridishModelBuilder()
model = Model(model_desc, droppath=False, affine=False)
summary(model, [64, 3, 32, 32])
exit(0)
# Copyright (c) Microsoft Corporation.
# Licensed under the MIT license.
import logging
from archai.datasets import data
from archai.common import utils
from archai.common.timing import MeasureTime, print_all_timings, print_timing, get_timing
from archai.common.common import logger, common_init
conf = common_init(config_filepath='confs/algos/darts.yaml',
param_args=['--common.experiment_name', 'restnet_test'])
conf_eval = conf['nas']['eval']
conf_loader = conf_eval['loader']
conf_loader['train_batch'] = 512
conf_loader['test_batch'] = 4096
conf_loader['cutout'] = 0
train_dl, _, test_dl = data.get_data(conf_loader)
@MeasureTime
def iter_dl(dl):
dummy = 0.0
for x, y in train_dl:
x = x.cuda()
y = y.cuda()
dummy += len(x)
# dummy += len(y)
return dummy
from archai.darts.darts_micro_builder import DartsMicroBuilder
from archai.common.common import common_init
from archai.nas.evaluate import eval_arch
if __name__ == '__main__':
conf = common_init(
config_filepath='confs/darts_cifar.yaml',
param_args=['--common.experiment_name', 'darts_cifar_eval'])
conf_eval = conf['nas']['eval']
# evaluate architecture using eval settings
eval_arch(conf_eval, micro_builder=DartsMicroBuilder())
exit(0)
import json
import os
from archai.common.common import logger, common_init, expdir_abspath
from archai.data_aug.train import train_and_eval
if __name__ == '__main__':
conf = common_init(config_filepath='confs/aug_train_cifar.yaml',
param_args=[
"--autoaug.loader.aug", "fa_reduced_cifar10",
"--common.experiment_name", "autoaug_train"
])
import time
t = time.time()
save_path = expdir_abspath('model.pth')
# result = train_and_eval(conf, val_ratio=conf['val_ratio'], val_fold=conf['val_fold'],
# save_path=save_path, only_eval=conf['only_eval'], metric='test')
# TODO: Will fail if val_ratio=0 since we are not using latest training infrastructure
# TODO: Move val_ratio, val_fold, metric to config file
result = train_and_eval(conf,
val_ratio=0.2,
val_fold=0,
save_path=save_path,
only_eval=False,
metric='test')
elapsed = time.time() - t
logger.info('training done.')
logger.info('model: %s' % conf['autoaug']['model'])
logger.info('augmentation: %s' % conf['autoaug']['loader']['aug'])
import torch
from archai import cifar10_models
from archai.common.trainer import Trainer
from archai.common.config import Config
from archai.common.common import common_init
from archai.datasets import data
def train_test(conf_eval: Config):
conf_loader = conf_eval['loader']
conf_trainer = conf_eval['trainer']
# create model
Net = cifar10_models.resnet34
model = Net().to(torch.device('cuda', 0))
# get data
data_loaders = data.get_data(conf_loader)
# train!
trainer = Trainer(conf_trainer, model)
trainer.fit(data_loaders)
if __name__ == '__main__':
conf = common_init(
config_filepath='confs/algos/resnet.yaml;confs/datasets/cifar100.yaml')
conf_eval = conf['nas']['eval']
train_test(conf_eval)
import torch
from archai import cifar10_models
from archai.common.trainer import Trainer
from archai.common.config import Config
from archai.common.common import common_init
from archai.datasets import data
def train_test(conf_eval: Config):
conf_loader = conf_eval['loader']
conf_trainer = conf_eval['trainer']
# create model
Net = cifar10_models.resnet34
model = Net().to(torch.device('cuda', 0))
# get data
train_dl, _, test_dl = data.get_data(conf_loader)
# train!
trainer = Trainer(conf_trainer, model)
trainer.fit(train_dl, test_dl)
if __name__ == '__main__':
conf = common_init(config_filepath='confs/algos/resnet.yaml')
conf_eval = conf['nas']['eval']
train_test(conf_eval)
def get_filepath(suffix):
conf = common_init(config_filepath='confs/algos/darts.yaml',
param_args=['--common.experiment_name', 'test_basename' + f'_{suffix}'
])
return utils.full_path(os.path.join('$expdir' ,'somefile.txt'))
from archai.common.common import common_init
from archai.darts.darts_micro_builder import DartsMicroBuilder
from archai.nas.evaluate import eval_arch
if __name__ == '__main__':
conf = common_init(
config_filepath='confs/imagenet_darts.yaml',
param_args=['--common.experiment_name', 'darts_imagenet_eval'])
conf_eval = conf['nas']['eval']
micro_builder = DartsMicroBuilder()
eval_arch(conf_eval, micro_builder=micro_builder)
exit(0)
from archai.common.common import common_init
from archai.random_arch.random_micro_builder import RandomMicroBuilder
from archai.nas import nas_utils
if __name__ == '__main__':
conf = common_init(config_filepath='confs/random_cifar.yaml',
param_args=['--common.experiment_name', 'random_cifar_search'])
# region config
conf_search = conf['nas']['search']
conf_model_desc = conf_search['model_desc']
final_desc_filename = conf_search['final_desc_filename']
# endregion
# create model and save it to yaml
# NOTE: there is no search here as the models are just randomly sampled
model_desc = nas_utils.create_macro_desc(conf_model_desc,
aux_tower=False,
template_model_desc=None)
macro_builder = RandomMicroBuilder()
macro_builder.build(model_desc, 0)
# save model to location specified by eval config
model_desc.save(final_desc_filename)
exit(0)
