def predict(self, configs, progress=None):
configs = ConfigSpec(configs)
opt_model = OptModel()
opt_model.add_child_spec(configs, 'configs')
zazu = ZaZu(opt_model, remote=True)
zazu.run_inference()
def search(self, configs, progress=None):
configs = ConfigSpec(configs)
opt_model = OptModel()
opt_model.add_child_spec(configs, 'configs')
zazu = ZaZu(opt_model, remote=True)
zazu.find_best_model()
zazu.hp_search()
checkpoint_paths_list = glob.glob('*checkpoint*.pt')
save_info = {
'package_name': self.package_name,
'execution_id': progress.execution.id
}
project_name = opt_model.dataloop['project']
project = dl.projects.get(project_name=project_name)
# model_name = opt_model.name
# model_obj = dl.models.get(model_name=model_name)
logger.info('uploading checkpoints.....')
for checkpoint_path in checkpoint_paths_list:
# model_obj.checkpoints.upload(checkpoint_name=checkpoint_path.split('.')[0], local_path=checkpoint_path)
project.artifacts.upload(filepath=checkpoint_path,
package_name=save_info['package_name'],
execution_id=save_info['execution_id'])
logger.info('finished uploading checkpoints')
def train(self, configs, progress=None):
configs = ConfigSpec(configs)
opt_model = OptModel()
opt_model.add_child_spec(configs, 'configs')
zazu = ZaZu(opt_model, remote=True)
zazu.train_new_model()
def search(self, configs, progress=None):
configs = ConfigSpec(configs)
opt_model = OptModel()
opt_model.add_child_spec(configs, 'configs')
zazu = ZaZu(opt_model, remote=True)
zazu.find_best_model()
zazu.hp_search()
dataset_obj = get_dataset_obj(configs['dataloop'])
id = dataset_obj.project.id
if args.search:
zazu_service.execute(function_name='search',
execution_input=inputs,
project_id=id)
if args.predict:
zazu_service.execute(function_name='predict',
execution_input=inputs,
project_id=id)
else:
logger = init_logging(__name__)
this_path = path = os.getcwd()
configs_path = os.path.join(this_path, 'configs.json')
configs = ConfigSpec(configs_path)
opt_model = OptModel()
opt_model.add_child_spec(configs, 'configs')
zazu = ZaZu(opt_model, remote=args.remote)
if args.search:
zazu.find_best_model()
zazu.hp_search()
if args.train:
zazu.train_new_model()
if args.predict:
zazu.run_inference()
if args.predict_once:
zazu.one_time_inference('/home/noam/0120122798.jpg',
'checkpoint0.pt')
parser.add_argument("--update", action='store_true', default=False)
parser.add_argument("--search", action='store_true', default=False)
parser.add_argument("--train", action='store_true', default=False)
parser.add_argument("--predict", action='store_true', default=False)
parser.add_argument("--zazu_timer", action='store_true', default=False)
parser.add_argument("--checkpoint_path", type=str, default='/home/noam/ZazuML/best_checkpoint.pt')
parser.add_argument("--dataset_path", type=str, default='')
parser.add_argument("--output_path", type=str, default='')
args = parser.parse_args()
with open('configs.json', 'r') as fp:
configs = json.load(fp)
logger = init_logging(__name__)
this_path = path = os.getcwd()
configs_path = os.path.join(this_path, 'configs.json')
configs = ConfigSpec('configs.json')
opt_model = OptModel('models.json')
opt_model.add_child_spec(configs, 'configs')
zazu = ZaZu(opt_model, remote=args.remote)
if args.search:
zazu.hp_search()
if args.train:
adapter = AdapterModel()
adapter.load(checkpoint_path=args.checkpoint_path)
adapter.train()
print('model checkpoint is saved to: ', adapter.checkpoint_path)
if args.predict:
predict(pred_on_path=args.dataset_path, output_path=args.output_path,
checkpoint_path=args.checkpoint_path, threshold=0.5)