def getOptions():
parser = argparse.ArgumentParser(description='Test data loader.', formatter_class=argparse.ArgumentDefaultsHelpFormatter)
malpiOptions.addMalpiOptions( parser )
args = parser.parse_args()
malpiOptions.preprocessOptions(args)
return args
early.on_train_end()
# finished, final model:
if save_model:
vae.save_json("tf_vae/vae.json")
return best_loss
if __name__ == "__main__":
import argparse
import malpiOptions
parser = argparse.ArgumentParser(description='Test data loader.', formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('--learning_rate', type=float, default=0.0001, help='Learning rate')
parser.add_argument('--z_size', type=int, default=32, help='Latent space size')
parser.add_argument('--batch_size', type=int, default=100, help='Mini-batch size')
parser.add_argument('--kl_tolerance', type=float, default=0.5, help='Max KL tolerance')
parser.add_argument('--epochs', type=int, default=100, help='Mini-batch size')
malpiOptions.addMalpiOptions( parser )
args = parser.parse_args()
malpiOptions.preprocessOptions(args)
if args.test_only:
runTests(args)
exit()
main( args.dirs, z_size=args.z_size, batch_size=args.batch_size, learning_rate=args.learning_rate, kl_tolerance=args.kl_tolerance, epochs=args.epochs, save_model=False, verbose=True )
help='number of episodes to use to train')
parser.add_argument('--new_model',
action='store_true',
help='start a new model from scratch?')
parser.add_argument('--epochs',
default=10,
help='number of epochs to train for')
parser.add_argument(
'--val',
help='File with one drive data directory per line for validation')
parser.add_argument('--val_split',
type=float,
default=0.2,
help='Percent validation split')
addMalpiOptions(parser)
args = parser.parse_args()
preprocessOptions(args)
if len(args.dirs) == 0 and not (args.test_only or args.start_batch):
parser.print_help()
print("\nNo directories supplied")
exit()
if args.val is None:
last = int(len(args.dirs) * args.val_split)
np.random.shuffle(args.dirs)
test = args.dirs[:last]
train = args.dirs[last:]
args.dirs = train
args.val = test
def getOptions():
parser = argparse.ArgumentParser(
description='Train on robot image/action data.',
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
#parser.add_argument('dirs', nargs='*', metavar="Directory", help='A directory containing recorded robot data')
#parser.add_argument('-f', '--file', help='File with one directory per line')
parser.add_argument(
'--fc',
action="store_true",
default=False,
help='Train a model with a fully connected layer (no RNN)')
parser.add_argument(
'--dk',
action="store_true",
default=False,
help='Train a model with DonkeyCar style Convolution layers')
parser.add_argument('--early',
action="store_true",
default=False,
help='Stop training early if learning plateaus')
parser.add_argument('--runs',
type=int,
default=1,
help='How many runs to train')
parser.add_argument('--name',
help='Display name for this training experiment')
parser.add_argument(
'--aux',
default=None,
help='Use this auxiliary data in place of standard actions')
parser.add_argument('--model',
default=None,
help='A file containing weights to pre-load the model')
parser.add_argument(
'--val',
default=None,
help='A file with a list of directories to be used for validation')
parser.add_argument('--val_split',
type=float,
default=0.2,
help='Percent validation split')
parser.add_argument(
'--notify',
help='Email address to notify when the training is finished')
#parser.add_argument('--test_only', action="store_true", default=False, help='run tests, then exit')
parser.add_argument('--random',
action="store_true",
default=False,
help='Test an untrained model, then exit')
parser.add_argument(
'--aug',
type=int,
default=None,
help=
'Augment images by this factor. 2 = twice as many images, half of which are altered'
)
malpiOptions.addMalpiOptions(parser)
args = parser.parse_args()
malpiOptions.preprocessOptions(args)
if len(args.dirs) == 0 and not args.test_only:
parser.print_help()
print("\nNo directories supplied")
exit()
if args.val is None:
last = int(len(args.dirs) * args.val_split)
np.random.shuffle(args.dirs)
test = args.dirs[:last]
train = args.dirs[last:]
args.dirs = train
args.val = test
return args