#!/usr/bin/env python
import math
import sys, argparse
import time
import numpy as np
import owl
import owl.net as net
from owl.net.trainer import NetTrainer
if __name__ == "__main__":
# parse command line arguments
parser = argparse.ArgumentParser()
parser.add_argument('solver_file', help='caffe solver configure file')
parser.add_argument('snapshot', help='the snapshot idx to start from', type=int, default=0)
parser.add_argument('num_gpu', help='number of gpus to use', type=int, default=1)
parser.add_argument('freq', help='frequency (number of minibatches) to call wait_for_all', type=int, default=1)
(args, remain) = parser.parse_known_args()
solver_file = args.solver_file
num_gpu = args.num_gpu
snapshot = args.snapshot
freq = args.freq
print ' === Training using %d gpus, start from snapshot %d === ' % (num_gpu, snapshot)
sys_args = [sys.argv[0]] + remain
trainer = NetTrainer(solver_file, snapshot, num_gpu, freq)
trainer.build_net()
trainer.run()
if __name__ == "__main__":
# parse command line arguments
parser = argparse.ArgumentParser()
parser.add_argument('solver_file', help='caffe solver configure file')
parser.add_argument('-n',
'--num_gpu',
help='number of gpus to use',
action='store',
type=int,
default=1)
parser.add_argument('--snapshot',
help='the snapshot idx to start from',
action='store',
type=int,
default=0)
(args, remain) = parser.parse_known_args()
solver_file = args.solver_file
num_gpu = args.num_gpu
snapshot = args.snapshot
print ' === Training using %d gpus, start from snapshot %d === ' % (
num_gpu, snapshot)
sys_args = [sys.argv[0]] + remain
owl.initialize(sys_args)
trainer = NetTrainer(solver_file, snapshot, num_gpu)
trainer.build_net()
trainer.run()
#!/usr/bin/env python
import math
import sys, argparse
import time
import numpy as np
import owl
import owl.net as net
from owl.net.trainer import NetTrainer
if __name__ == "__main__":
# parse command line arguments
parser = argparse.ArgumentParser()
parser.add_argument('solver_file', help='caffe solver configure file')
parser.add_argument('checklayer', help='the layer to be checked')
parser.add_argument('snapshot', help='the snapshot idx to start from', type=int, default=0)
(args, remain) = parser.parse_known_args()
solver_file = args.solver_file
checklayer = args.checklayer
snapshot = args.snapshot
print ' === Gradient Check, start from snapshot %d === ' % (snapshot)
sys_args = [sys.argv[0]] + remain
trainer = NetTrainer(solver_file, snapshot, 1)
trainer.build_net()
trainer.gradient_checker(checklayer)