def main():
parser = argparse.ArgumentParser()
arg = parser.add_argument
arg('--model_name', type=str, default='mask_512', help='String model name from models dictionary')
arg('--seed', type=int, default=1234, help='Random seed')
arg('--fold', type=int, default=0, help='Validation fold')
arg('--weights_dir', type=str, default='', help='Directory for loading model weights')
arg('--epochs', type=int, default=12, help='Current epoch')
arg('--lr', type=float, default=1e-3, help='Initial learning rate')
arg('--debug', type=bool, default=False, help='If the debugging mode')
args = parser.parse_args()
set_seed(args.seed)
# get data
if ON_SERVER:
level5data = LyftDataset(data_path = '.', json_path='../../input/train_data', verbose=True) # server
else:
level5data = LyftDataset(data_path = '../input/', json_path='../input/train_data', verbose=True) # local laptop
classes = ["car", "motorcycle", "bus", "bicycle", "truck", "pedestrian", "other_vehicle", "animal", "emergency_vehicle"]
# "bev" folder
data_folder = os.path.join(OUTPUT_ROOT, "bev_data")
# choose model
model = get_maskrcnn_model(NUM_CLASSES)
checkpoint= f'{OUTPUT_ROOT}/checkpoints/'
train(model, model_name='mask_512', data_folder=data_folder,
level5data = level5data,
fold=args.fold, debug=args.debug, img_size=IMG_SIZE, bev_shape=BEV_SHAPE,
epochs=args.epochs, batch_size=16, num_workers=4,
learning_rate = args.lr, resume_weights=args.weights_dir, resume_epoch=0)
def _setup(self, variant):
set_seed(variant['run_params']['seed'])
self._variant = variant
gpu_options = tf.GPUOptions(allow_growth=True)
session = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options))
tf.keras.backend.set_session(session)
self._session = tf.keras.backend.get_session()
self.train_generator = None
self._built = False
def _setup(self, params):
self._params = params
#### set up tf session
set_seed(params['run_params']['seed'])
gpu_options = tf.GPUOptions(allow_growth=True)
session = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options))
tf.keras.backend.set_session(session)
self._session = tf.keras.backend.get_session()
self.train_generator = None
self._built = False
def main():
parser = argparse.ArgumentParser()
arg = parser.add_argument
arg('--model_name', type=str, default='mask_512', help='String model name from models dictionary')
arg('--seed', type=int, default=1234, help='Random seed')
arg('--fold', type=int, default=0, help='Validation fold')
arg('--weights_dir', type=str, default='', help='Directory for loading model weights')
arg('--epochs', type=int, default=12, help='Current epoch')
arg('--lr', type=float, default=1e-3, help='Initial learning rate')
arg('--debug', type=bool, default=False, help='If the debugging mode')
args = parser.parse_args()
set_seed(args.seed)
classes = ["car", "motorcycle", "bus", "bicycle", "truck", "pedestrian", "other_vehicle", "animal", "emergency_vehicle"]
model = get_unet_twohead_model(encoder='resnet18', num_classes=len(classes)+1)
data_folder = os.path.join(OUTPUT_ROOT, "bev_data")
checkpoint= f'{OUTPUT_ROOT}/checkpoints/unet_resnet152_384_fold_2/unet_resnet152_384_fold_2_epoch_6.pth'
train(model, model_name=args.model_name, data_folder=data_folder,
fold=args.fold, debug=args.debug, img_size=IMG_SIZE, learning_rate= args.lr,
epochs=args.epochs, batch_size=32, num_workers=4, resume_weights='', resume_epoch=0)
import os
import time
from tqdm import tqdm
from graphino.GCN.GCN_model import GCN
# Training settings
from graphino.training import evaluate, train_epoch, get_dataloaders, get_dirs
from utilities.hyperparams_and_args_GCN import get_argparser
from utilities.utils import set_seed
from utilities.model_logging import update_tqdm, save_model
from utilities.optimization import get_optimizer, get_loss
params, net_params = get_argparser()
set_seed(params['seed'])
(adj, static_feats, _), (trainloader, valloader,
testloader) = get_dataloaders(params, net_params)
ckpt_dir, log_dir = get_dirs(params, net_params)
# Model and optimizer
model = GCN(net_params, static_feat=static_feats, adj=adj)
optimizer = get_optimizer(params['optimizer'],
model,
lr=params['lr'],
weight_decay=params['weight_decay'],
nesterov=params['nesterov'])
criterion = get_loss(params['loss'])
# Train model
device = 'cuda'
t_total = time.time()
if torch.cuda.is_available():
if not use_cuda:
print("\n### WARNING: You have a CUDA device, "
"so you should probably run with --cuda\n")
if use_cuda:
print("Using GPU device_id:0")
device_id = 0
utils.use_cuda()
torch.cuda.set_device(device_id)
else:
print("Cuda Not Available!!")
device_id = None
utils.set_seed(seed)
ckptfilename = getCkptName(args.ckptname)
ckptpath = os.path.join(ckptroot, modeltype, ckptfilename)
print("CKPT PATH: {}".format(ckptpath))
# Initialized reader, model and optimizer
trainer = Trainer()
print("Done modelinit")
print("Mode : {}".format(mode))
# print(trainer.tr_reader.ans2idx)
if mode == 'train':
(bestmodel, bestval, beststeps, steps) = trainer.train()
print("Saving model: {}".format(ckptpath))
utils.save_checkpoint(m=bestmodel,