'unsupervised pre-training.')
parser.add_argument('--batch',
dest='batchSize',
type=int,
default=1,
help='Batch size for training and test sets.')
parser.add_argument('--base',
dest='base',
type=str,
default='./leNet5',
help='Base name of the network output and temp files.')
parser.add_argument('image', help='Input image to train.')
options = parser.parse_args()
# setup the logger
log = setupLogging('likenessFinder', options.level, options.logfile)
# read the file into memory
if log is not None:
log.info('Reading the Input into Memory...')
imageFromDisk = Image.open(options.image)
imageFromDisk.load()
# decimate everything for speed
thumbOrig = imageFromDisk.copy()
thumbOrig.thumbnail((imageFromDisk.size[0] / options.scale,
imageFromDisk.size[1] / options.scale),
Image.ANTIALIAS)
# load a network from disk or train a new network
network = createNetwork(thumbOrig, log)
help='Synapse for the deep network to distill. This ' +
'network should be trained and ready.')
parser.add_argument('--syn',
dest='synapse',
type=str,
default=None,
help='Load from a previously saved network.')
parser.add_argument('data',
help='Directory or pkl.gz file for the ' +
'training and test sets. This can be ' +
'the location of a dark pickle.')
options = parser.parse_args()
# setup the logger
logName = 'distillery: ' + options.data
log = setupLogging(logName, options.level, options.logfile)
prof = Profiler(log=log, name=logName, profFile=options.profile)
# create a random number generator for efficiency
from numpy.random import RandomState
from operator import mul
from time import time
rng = RandomState(int(time()))
#rng = RandomState(4567) # always initialize the same
# NOTE: The pickleDataset will silently use previously created pickles if
# one exists (for efficiency). So watch out for stale pickles!
# NOTE: User may pass a dark pickle into here, and the logic will react
# appropriately to the situation.
train, test, labels = ingestImagery(filepath=options.data,
shared=True,
parser.add_argument('--hidden', dest='hidden', type=int, default=400,
help='Number of Neurons in Hidden Layer.')
parser.add_argument('--neuron', dest='neuron', type=int, default=250,
help='Number of Neurons in Output Layer.')
parser.add_argument('--epoch', dest='numEpochs', type=int, default=15,
help='Number of epochs to run per layer during ' +
'unsupervised pre-training.')
parser.add_argument('--batch', dest='batchSize', type=int, default=1,
help='Batch size for training and test sets.')
parser.add_argument('--base', dest='base', type=str, default='./leNet5',
help='Base name of the network output and temp files.')
parser.add_argument('image', help='Input image to train.')
options = parser.parse_args()
# setup the logger
log = setupLogging('likenessFinder', options.level, options.logfile)
# read the file into memory
if log is not None :
log.info('Reading the Input into Memory...')
imageFromDisk = Image.open(options.image)
imageFromDisk.load()
# decimate everything for speed
thumbOrig = imageFromDisk.copy()
thumbOrig.thumbnail((imageFromDisk.size[0] / options.scale,
imageFromDisk.size[1] / options.scale),
Image.ANTIALIAS)
# load a network from disk or train a new network
network = createNetwork(thumbOrig, log)
parser.add_argument('--numChips', dest='numChips', type=int, default=1000,
help='Number of epochs to run per layer.')
parser.add_argument('--epoch', dest='numEpochs', type=int, default=15,
help='Number of epochs to run per layer.')
parser.add_argument('--batch', dest='batchSize', type=int, default=100,
help='Batch size for training and test sets.')
parser.add_argument('--base', dest='base', type=str, default='./leNet5',
help='Base name of the network output and temp files.')
parser.add_argument('--syn', dest='synapse', type=str, default=None,
help='Load from a previously saved network.')
parser.add_argument('data', help='Directory or pkl.gz file for the ' +
'training and test sets')
options = parser.parse_args()
# setup the logger
log = setupLogging('cnnPreTrainer: ' + options.data,
options.level, options.logfile)
# NOTE: The pickleDataset will silently use previously created pickles if
# one exists (for efficiency). So watch out for stale pickles!
chipArgs = {'chipSize': options.chipSize,
'numChips': options.numChips}
train = ingestImagery(filepath=options.data, shared=True,
batchSize=options.batchSize,
log=log, chipFunc=randomChip, kwargs=chipArgs)
if options.synapse is not None :
# load a previously saved network
network = StackedAENetwork(train, log=log)
network.load(options.synapse)
else :
network = buildStackedAENetwork(
help='Synapse for the shallow target network. This ' +
'network should be populated with freshly ' +
'initialized layers for optimal results.')
parser.add_argument('--deep', dest='deep', type=str, default=None,
help='Synapse for the deep network to distill. This ' +
'network should be trained and ready.')
parser.add_argument('--dark', dest='dark', type=str, default=None,
help='pkl.gz file previously created by the ' +
'distillery for dark knowledge transfer.')
parser.add_argument('--data', dest='data', type=str, default=None,
help='Directory or pkl.gz file for the training and ' +
'test sets')
options = parser.parse_args()
# setup the logger
log = setupLogging('distillery: ' + options.data,
options.level, options.logfile)
# if the user specified a deep network and dataset, then distill the
# knowledge into a new pickle to use for training.
if options.deep is not None :
from dataset.ingest.distill import distillKnowledge
if options.dark is not None :
raise Exception('Only specify one usage, --deep or --dark.')
if options.data is None :
raise Exception('When specifying a deep network, please also ' +
'specify a dataset to distill using --data.')
# distill knowledge out of the deep network into a pickle
deepNet = ClassifierNetwork(filepath=options.deep, log=log)
options.dark = distillKnowledge(deepNet=deepNet,
help='Number of inferior validation checks to end ' +
'the supervised learning session.')
parser.add_argument('--holdout', dest='holdout', type=float, default=.05,
help='Percent of data to be held out for testing.')
parser.add_argument('--batch', dest='batchSize', type=int, default=5,
help='Batch size for training and test sets.')
parser.add_argument('--base', dest='base', type=str, default='./leNet5',
help='Base name of the network output and temp files.')
parser.add_argument('--syn', dest='synapse', type=str, default=None,
help='Load from a previously saved network.')
parser.add_argument('data', help='Directory or pkl.gz file for the ' +
'training and test sets')
options = parser.parse_args()
# setup the logger
log = setupLogging('semiSupervisedTrainer: ' + options.data,
options.level, options.logfile)
# create a random number generator for efficiency
from numpy.random import RandomState
from operator import mul
rng = RandomState(int(time()))
# NOTE: The pickleDataset will silently use previously created pickles if
# one exists (for efficiency). So watch out for stale pickles!
train, test, labels = ingestImagery(filepath=options.data, shared=False,
batchSize=options.batchSize,
holdoutPercentage=options.holdout,
log=log)
trainShape = train[0].shape
# create the stacked network -- LeNet-5 (minus the output layer)