def main(expdir, recipe, computing):
'''main function'''
overwrite = False
if os.path.isdir(expdir):
text = ''
while text not in ('o', 'r'):
text = raw_input('%s already exists, do you want to '
'resume experiment (r) or overwrite (o) '
'(respond with o or r)' % expdir)
if text == 'o':
overwrite = True
else:
#create the experiments directory
os.makedirs(expdir)
#copy the config files
if overwrite:
shutil.copyfile(os.path.join(recipe, 'acquisition.cfg'),
os.path.join(expdir, 'acquisition.cfg'))
else:
tools.safecopy(os.path.join(recipe, 'acquisition.cfg'),
os.path.join(expdir, 'acquisition.cfg'))
shutil.copyfile(os.path.join(recipe, 'coder.cfg'),
os.path.join(expdir, 'coder.cfg'))
shutil.copyfile(os.path.join(recipe, 'structure.xml'),
os.path.join(expdir, 'structure.xml'))
shutil.copyfile(os.path.join(recipe, 'coder.cfg'),
os.path.join(expdir, 'coder.cfg'))
shutil.copyfile(os.path.join(recipe, 'structure.xml'),
os.path.join(expdir, 'structure.xml'))
shutil.copyfile(os.path.join(recipe, 'database.cfg'),
os.path.join(expdir, 'database.cfg'))
shutil.copyfile(os.path.join(recipe, 'cross_validation_ppall.cfg'),
os.path.join(expdir, 'cross_validation_ppall.cfg'))
acquisitionconf = ConfigParser()
acquisitionconf.read(os.path.join(recipe, 'acquisition.cfg'))
modelname = acquisitionconf.get('acquisition', 'name')
shutil.copyfile(
os.path.join(os.getcwd(), 'assist', 'acquisition', 'defaults',
modelname + '.cfg'),
os.path.join(expdir, modelname + '.cfg'))
#read the cross_validation config file
expconf = ConfigParser()
expconf.read(os.path.join(recipe, 'cross_validation.cfg'))
#default conf file
default = os.path.join(os.path.dirname(os.path.realpath(__file__)),
'defaults', 'cross_validation.cfg')
#apply the defaults
if os.path.exists(default):
tools.default_conf(expconf, default)
expconf = dict(expconf.items('cross_validation'))
#read the data config file
if not os.path.exists(os.path.join(recipe, 'database.cfg')):
raise Exception('cannot find database.cfg in %s' % recipe)
dataconf = ConfigParser()
dataconf.read(os.path.join(recipe, 'database.cfg'))
#read the coder config file
coderconf = ConfigParser()
coderconf.read(os.path.join(expdir, 'coder.cfg'))
# filter out all speakers with less than 100 examples
# (in the FluentSpeechCommands dataset (~20%))
bad_spks = []
if os.path.exists(os.path.join(recipe, 'FS_linecounts.txt')):
for l in open(os.path.join(recipe, 'FS_linecounts.txt')):
splitline = l.strip().split(' ')
if int(splitline[1]) < 100:
bad_spks.append(splitline[0])
print bad_spks
for speaker in dataconf.sections():
if speaker in bad_spks:
continue
print 'speaker: %s' % (speaker)
#create the speaker directory
if os.path.isdir(os.path.join(expdir, speaker)):
if overwrite:
shutil.rmtree(os.path.join(expdir, speaker))
os.makedirs(os.path.join(expdir, speaker))
else:
os.makedirs(os.path.join(expdir, speaker))
#create a task structure file
structure = Structure(os.path.join(expdir, 'structure.xml'))
#create a coder
coder = coder_factory.factory(coderconf.get('coder',
'name'))(structure,
coderconf)
#read and code all the tasks
labelvecs = []
names = []
taskstrings = dict()
for line in open(dataconf.get(speaker, 'tasks')):
splitline = line.strip().split(' ')
name = speaker + '_' + splitline[0]
names.append(name)
taskstring = ' '.join(splitline[1:])
taskstrings[name] = taskstring
task = read_task(taskstring)
labelvecs.append(coder.encode(task))
#devide the data into blocks
blocksfile = os.path.join(expdir, speaker, 'blocks.pkl')
if os.path.exists(blocksfile):
with open(blocksfile, 'rb') as fid:
blocks = pickle.load(fid)
else:
blocks = make_blocks(np.array(labelvecs), expconf,
dataconf.get(speaker, 'features'))
with open(blocksfile, 'wb') as fid:
pickle.dump(blocks, fid)
#create train-testsets for all experiments
#seed the random number generator
random.seed(3105)
trainids = [None] * (len(blocks) - 1)
testids = [None] * (len(blocks) - 1)
for b in range(len(blocks) - 1):
trainids[b] = [None] * int(expconf['numexp'])
testids[b] = [None] * int(expconf['numexp'])
for e in range(int(expconf['numexp'])):
trainids[b][e] = list(
itertools.chain.from_iterable(random.sample(blocks,
b + 1)))
testids[b][e] = [
x for x in range(len(names)) if x not in trainids[b][e]
]
#read the feature files
features = dict()
for l in open(os.path.join(dataconf.get(speaker, 'features'),
'feats')):
splitline = l.strip().split(' ')
featname = speaker + '_' + splitline[0]
features[featname] = ' '.join(splitline[1:])
#create an expdir for each experiment
b = int(expconf['startblocks']) - 1
while True:
for e in range(int(expconf['numexp'])):
print ' train blocks: %d, experiment %s' % (b + 1, e)
#creat the directory
subexpdir = os.path.join(expdir, speaker,
'%dblocks_exp%d' % (b + 1, e))
if os.path.exists(os.path.join(subexpdir, 'f1')):
continue
if not os.path.isdir(subexpdir):
os.makedirs(subexpdir)
#create pointers to the config files
tools.symlink(os.path.join(expdir, 'acquisition.cfg'),
os.path.join(subexpdir, 'acquisition.cfg'))
tools.symlink(os.path.join(expdir, 'coder.cfg'),
os.path.join(subexpdir, 'coder.cfg'))
tools.symlink(os.path.join(expdir, 'structure.xml'),
os.path.join(subexpdir, 'structure.xml'))
tools.symlink(os.path.join(expdir, 'database.cfg'),
os.path.join(subexpdir, 'database.cfg'))
if not os.path.exists(os.path.join(subexpdir, 'trainfeats')):
trainutts = [names[i] for i in trainids[b][e]]
print 'number of examples: %d' % len(trainutts)
testutts = [names[i] for i in testids[b][e]]
#create the train and test sets
tools.writefile(os.path.join(subexpdir, 'trainfeats'),
{utt: features[utt]
for utt in trainutts})
tools.writefile(
os.path.join(subexpdir, 'traintasks'),
{utt: taskstrings[utt]
for utt in trainutts})
tools.writefile(os.path.join(subexpdir, 'testfeats'),
{utt: features[utt]
for utt in testutts})
tools.writefile(
os.path.join(subexpdir, 'testtasks'),
{utt: taskstrings[utt]
for utt in testutts})
if computing in ('condor', 'condor_gpu'):
#create the outputs directory
if not os.path.isdir(os.path.join(subexpdir, 'outputs')):
os.makedirs(os.path.join(subexpdir, 'outputs'))
if computing == 'condor_gpu':
jobfile = 'run_script_GPU.job'
else:
jobfile = 'run_script.job'
#only submit the job if it not running yet
in_queue = os.popen(
'if condor_q -nobatch -wide | grep -q %s; '
'then echo true; else echo false; fi' %
subexpdir).read().strip() == 'true'
#submit the condor job
if not in_queue:
os.system('condor_submit expdir=%s script=train_test'
' assist/condor/%s' % (subexpdir, jobfile))
else:
train_test.main(subexpdir)
newb = (b + 1) * int(expconf['scale']) + int(
expconf['increment']) - 1
newb = min(newb, len(blocks) - 2)
if b == newb:
break
else:
b = newb
def execute(node, previous, experiment_folder):
"""
Execute a task defined by the given node in the experiment graph.
Parameters
----------
node : Element
The node to be executed.
previous : dict (or list of dict)
Dictionary of the experiment's running-time variables after the
end of the parent node's execution.
May be a list of dictionaries in the special case of a fusion node,
which has more than one parent.
experiment_folder : string
String with the path to the experiment folder, where the files of the
experiment will be saved.
Returns
-------
exp_param : dict
The updated dictionary of the experiment's running-time variables after
the node's execution.
"""
global execution_time
global tex_path
global tex_dict
global openset_experiment
exp_param = previous
parameters = ast.literal_eval(node.get("parameters"))
node_id = node.attrib['id']
#Get node name
node_name = node.get('name')
if node.tag == "collection":
print "Collection", exp_param.keys()
images, classes, extract_path, read_time = \
read_collection.main(node_name, openset_experiment, parameters,
node_id)
execution_time += read_time
exp_param['images'] = images
exp_param['classes'] = classes
exp_param['extract_path'] = extract_path
elif node.tag == "train_test_method":
print "train_test_method", exp_param.keys()
images = exp_param['images']
classes = exp_param['classes']
images, classes, train_test_list, train_test_time = \
train_test.main(images, classes, experiment_folder, node_name,
parameters, openset_experiment, node_id)
execution_time += train_test_time
exp_param['images'] = images
exp_param['classes'] = classes
exp_param['train_test_list'] = train_test_list
exp_param['train_test_method'] = node_name
exp_param['train_test_parameters'] = parameters
elif node.tag == "descriptor":
print "descriptor", exp_param.keys()
images = exp_param['images']
extract_path = exp_param['extract_path']
classes_keys = exp_param['classes'].keys()
if node_name == "bag":
train_test_list = exp_param['train_test_list']
images, extract_time = extract_bag.main(images, train_test_list,
extract_path,
experiment_folder,
parameters, node_id)
elif node_name == "bovg":
train_test_list = exp_param['train_test_list']
images, extract_time = extract_bovg.main(images, train_test_list,
extract_path,
experiment_folder,
parameters, node_id)
else:
images, extract_time = extract_features.main(
images, classes_keys, extract_path, node_name, parameters,
node_id)
execution_time += extract_time
exp_param['images'] = images
exp_param['descriptor'] = node_name
elif node.tag == "normalizer":
try:
manager = Manager()
images = manager.dict(exp_param['images'])
train_test_list = exp_param['train_test_list']
except:
print "\n\tMissing Input. Exiting."
sys.exit(1)
norm_fv_paths, normalize_time = normalize_features.main(
images, train_test_list, experiment_folder, node_name, parameters,
node_id)
execution_time += normalize_time
del exp_param['images']
exp_param['fv_paths'] = norm_fv_paths
elif node.tag == "classifier":
try:
classes = exp_param['classes']
train_test_list = exp_param['train_test_list']
descriptor = exp_param['descriptor']
try:
fv_paths = exp_param['fv_paths']
del exp_param['fv_paths']
except:
images = exp_param['images']
fv_paths = util.save_file_extract(images, train_test_list,
experiment_folder)
except:
print "\n\tMissing Input. Exiting."
sys.exit(1)
images, classes_list, classify_time = classify.main(
fv_paths, classes.keys(), train_test_list, experiment_folder,
node_name, parameters, descriptor, node_id)
execution_time += classify_time
exp_param['images'] = images
exp_param['classes_list'] = classes_list
elif node.tag == "fusion_method":
len_exp_param = len(exp_param)
#list with the images dictionaries, classes dictionaries, and train and
# test set list
list_images = []
list_classes = []
list_train_test = []
extract_path = exp_param[INDEX_ZERO]['extract_path']
for index in range(len_exp_param):
try:
list_images.append(exp_param[index]['images'])
except:
images = {}
for fv_path in exp_param[index]['fv_paths']:
print "fv_path:", fv_path
images_new = util.read_fv_file(fv_path)
images = util.merge_dict(images, images_new)
list_images.append(images)
list_classes.append(exp_param[index]['classes'])
#In case that it performs the fusion of collections, there is no
# train_test_list
try:
list_train_test.append(exp_param[index]['train_test_list'])
except:
list_train_test.append(None)
#classes_list is present only after the classification module
try:
classes_list = exp_param[INDEX_ZERO]['classes_list']
except:
classes_list = None
try:
train_test_method = exp_param[INDEX_ZERO]['train_test_method']
train_test_parameters = exp_param[INDEX_ZERO][
'train_test_parameters']
except:
train_test_method = None
train_test_parameters = None
images, classes, train_test_list, fusion_time = \
fusion.main(list_images, list_classes, list_train_test,
classes_list, experiment_folder, node_name, parameters,
node_id)
execution_time += fusion_time
exp_param = {}
exp_param['images'] = images
exp_param['classes'] = classes
if train_test_list is not None:
exp_param['train_test_list'] = train_test_list
if classes_list is not None:
exp_param['classes_list'] = classes_list
if train_test_method is not None:
exp_param['train_test_method'] = train_test_method
exp_param['train_test_parameters'] = train_test_parameters
exp_param['descriptor'] = None
exp_param['extract_path'] = extract_path
elif node.tag == "evaluation_measure":
try:
images = exp_param['images']
train_test_list = exp_param['train_test_list']
classes_list = exp_param['classes_list']
except:
print "\n\tMissing Input. Exiting."
sys.exit(1)
evaluation_time, evaluation_path = evaluation.main(
images, train_test_list, classes_list, experiment_folder,
node_name, parameters, node_id)
execution_time += evaluation_time
#Dictionaries to create the tex file
train_test_method = exp_param['train_test_method']
train_test_parameters = str(exp_param['train_test_parameters'])
if train_test_method not in tex_dict:
tex_dict[train_test_method] = {}
train_test_dict = tex_dict[train_test_method]
if train_test_parameters not in train_test_dict:
train_test_dict[train_test_parameters] = {}
output_dict = train_test_dict[train_test_parameters]
if node_name not in output_dict:
output_dict[node_name] = []
list_output = [evaluation_path, classes_list[0], node_id]
if list_output not in output_dict[node_name]:
output_dict[node_name].append(list_output)
train_test_dict[train_test_parameters] = output_dict
tex_dict[train_test_method] = train_test_dict
elif node.tag == "preprocessing":
images = exp_param['images']
classes = exp_param['classes']
images, classes, preprocessing_time = preprocessing.main(
images, classes, experiment_folder, node_name, parameters, node_id)
execution_time += preprocessing_time
exp_param['images'] = images
exp_param['classes'] = classes
else:
print "Error. Unknown Tag."
sys.exit(1)
return exp_param
Augment each image to create more data
'''
i = 0
for batch in self.datagen.flow_from_directory(
directory=self.import_path,
save_to_dir=self.export_path,
save_prefix='keras_',
save_format='jpg',
batch_size=1, color_mode='rgb'):
i += 1
if i == self.num_images:
break
if __name__ == "__main__":
main()
os.chdir(home)
shape = (299, 299, 3)
import_path = 'data/Keras_Images'
export_path = 'data/Keras_Images'
#Save resized training and test photos into each folder
for i in artist:
#Test Images
test_path = 'data/Test/{0}/'.format(i)
test_resize = ImagePipeline(test_dict[i],
shape,
home,
import_path,
export_path)
test_resize.image_folder(test_path)
test_resize.save_folder('data/Test', i)
def main(expdir, recipe, computing):
'''main function'''
overwrite = False
if os.path.isdir(expdir):
text = ''
while text not in ('o', 'r'):
text = raw_input('%s already exists, do you want to '
'resume experiment (r) or overwrite (o) '
'(respond with o or r)' % expdir)
if text == 'o':
overwrite = True
else:
#create the experiments directory
os.makedirs(expdir)
#copy the config files
if overwrite:
shutil.copyfile(os.path.join(recipe, 'acquisition.cfg'),
os.path.join(expdir, 'acquisition.cfg'))
else:
tools.safecopy(os.path.join(recipe, 'acquisition.cfg'),
os.path.join(expdir, 'acquisition.cfg'))
shutil.copyfile(os.path.join(recipe, 'coder.cfg'),
os.path.join(expdir, 'coder.cfg'))
shutil.copyfile(os.path.join(recipe, 'structure.xml'),
os.path.join(expdir, 'structure.xml'))
shutil.copyfile(os.path.join(recipe, 'database.cfg'),
os.path.join(expdir, 'database.cfg'))
shutil.copyfile(os.path.join(recipe, 'cross_validation_ppall.cfg'),
os.path.join(expdir, 'cross_validation_ppall.cfg'))
acquisitionconf = ConfigParser()
acquisitionconf.read(os.path.join(recipe, 'acquisition.cfg'))
modelname = acquisitionconf.get('acquisition', 'name')
shutil.copyfile(
os.path.join(os.getcwd(), 'assist', 'acquisition', 'defaults',
modelname + '.cfg'),
os.path.join(expdir, modelname + '.cfg'))
#read the cross_validation config file
expconf = ConfigParser()
expconf.read(os.path.join(recipe, 'cross_validation_ppall.cfg'))
#default conf file
default = os.path.join(os.path.dirname(os.path.realpath(__file__)),
'defaults', 'cross_validation_ppall.cfg')
#apply the defaults
if os.path.exists(default):
tools.default_conf(expconf, default)
expconf = dict(expconf.items('cross_validation_ppall'))
#read the data config file
if not os.path.exists(os.path.join(recipe, 'database.cfg')):
raise Exception('cannot find database.cfg in %s' % recipe)
dataconf = ConfigParser()
dataconf.read(os.path.join(recipe, 'database.cfg'))
#read the coder config file
coderconf = ConfigParser()
coderconf.read(os.path.join(expdir, 'coder.cfg'))
# for word specific thresholds (not used anymore)
#if os.path.isfile(os.path.join(recipe,'word_thresholds.pkl')):
# print('File with wordthresholds found in recipe')
# shutil.copyfile(os.path.join(recipe, 'word_thresholds.pkl'),
# os.path.join(expdir, 'word_thresholds.pkl'))
# thresholdsarepresent = True
#else:
# print('No file found with wordthresholds, using a fixed one')
# thresholdsarepresent = False
labelvecs = []
names = []
taskstrings = dict()
features = dict()
print 'Searching for all speakers...'
for speaker in dataconf.sections():
print ' speaker: %s' % (speaker)
#create a task structure file
structure = Structure(os.path.join(expdir, 'structure.xml'))
#create a coder
coder = coder_factory.factory(coderconf.get('coder',
'name'))(structure,
coderconf)
# typesplit_coder.py line 51 to see all labels and corresponding output capsule numbers
#read and code all the tasks
for line in open(
dataconf.get(speaker, 'tasks')
): #'recording1_Voice_10 <move_rel direction="forward" distance="little" throttle="fast" />'
splitline = line.strip().split(' ')
name = speaker + '_' + splitline[0] #'recording1_Voice_10'
names.append(name)
taskstring = ' '.join(
splitline[1:]
) #'<move_rel_direction="forwqrd" distance="little" throttle="fast"/>'
taskstrings[name] = taskstring
task = read_task(taskstring)
labelvecs.append(coder.encode(task))
# read the feature files
for l in open(os.path.join(dataconf.get(speaker, 'features'),
'feats')):
splitline = l.strip().split(
' '
) #['recording1_Voice_10', '/esat/spchtemp/scratch/r0580562/databases/grabo_features/pp2/recording1_Voice_10.npy']
featname = speaker + '_' + splitline[0]
features[featname] = ' '.join(splitline[1:])
print 'Devide data into blocks...'
#devide the data into blocks, look for existing blocksfile in recipe because takes a very long time to make!!!
blocksfile = os.path.join(recipe, 'blocks.pkl')
if os.path.exists(blocksfile):
print 'Loading found blocks file (check if number of blocks is still the same)'
with open(blocksfile, 'rb') as fid:
blocks = pickle.load(fid)
else:
print 'No blocksfile found in recipe, making new one'
blocks = make_blocks(
np.array(labelvecs), expconf, expdir
) #massive list, matrix of [[..,..,..,..],[..,... .]] numbers between 1-350 approx
with open(blocksfile, 'wb') as fid:
pickle.dump(blocks, fid)
print 'Shuffle speakers...'
# look for existing train and test sets and load them in ('saved_ids' in recipe), because takes a very long time to make!!!
sets_properties = {}
if os.path.isdir(os.path.join(recipe, 'saved_ids')):
saved_ids = ConfigParser()
saved_ids.read(
os.path.join(recipe, 'saved_ids', 'cross_validation_ppall.cfg'))
sets_properties = dict(saved_ids.items('cross_validation_ppall'))
else:
sets_properties['numblocks'] = 0
sets_properties['numexp'] = 0
if (sets_properties['numblocks']
== expconf['numblocks']) and (sets_properties['numexp']
== expconf['numexp']):
print ' Loading found test recipe'
trainids_saved = os.path.join(recipe, 'saved_ids', 'trainids.pkl')
with open(trainids_saved, 'rb') as fid:
trainids = pickle.load(fid)
testids_saved = os.path.join(recipe, 'saved_ids', 'testids.pkl')
with open(testids_saved, 'rb') as fid:
testids = pickle.load(fid)
else:
print ' No saved test and train sets found with same crossvalidation configuration in the recipe'
# seed the random number generator
random.seed(3105)
trainids = [None] * (len(blocks) - 1) #len(blocks)=15
testids = [None] * (len(blocks) - 1)
print ' Number of blocks: %d' % (len(blocks))
b = 0
while b < (len(blocks) - 1):
#for b in range(len(blocks)-1):
print ' block %d' % b
trainids[b] = [None] * int(expconf['numexp'])
testids[b] = [None] * int(expconf['numexp'])
for e in range(int(expconf['numexp'])):
trainids[b][e] = list(
itertools.chain.from_iterable(random.sample(blocks,
b + 1)))
testids[b][e] = [
x for x in range(len(names)) if x not in trainids[b][e]
]
#scale factor to use more smaller blocks and less bigger blocks (for the curve, it saturates)
newb = int(
np.floor((b + 1) * float(expconf['scale']) +
int(expconf['increment']) - 1))
newb = min(newb, len(blocks) - 2)
if b == newb:
break
else:
b = newb
os.makedirs(os.path.join(expdir, 'saved_ids'))
trainids_saved = os.path.join(expdir, 'saved_ids', 'trainids.pkl')
testids_saved = os.path.join(expdir, 'saved_ids', 'testids.pkl')
with open(trainids_saved, 'wb') as fid:
pickle.dump(trainids, fid)
with open(testids_saved, 'wb') as fid:
pickle.dump(testids, fid)
shutil.copyfile(
os.path.join(recipe, 'cross_validation_ppall.cfg'),
os.path.join(expdir, 'saved_ids', 'cross_validation_ppall.cfg'))
#create an expdir for each experiment
b = int(expconf['startblocks']) - 1 #0
print 'Launch the experiments...'
while True:
for e in range(int(expconf['numexp'])):
print ' train blocks: %d, experiment %s' % (b + 1, e)
#creat the directory
subexpdir = os.path.join(expdir, '%dblocks_exp%d' % (b + 1, e))
if os.path.exists(os.path.join(subexpdir, 'f1')):
continue
if not os.path.isdir(subexpdir):
os.makedirs(subexpdir)
#create pointers to the config files
tools.symlink(os.path.join(expdir, 'acquisition.cfg'),
os.path.join(subexpdir, 'acquisition.cfg'))
tools.symlink(os.path.join(expdir, 'coder.cfg'),
os.path.join(subexpdir, 'coder.cfg'))
tools.symlink(os.path.join(expdir, 'structure.xml'),
os.path.join(subexpdir, 'structure.xml'))
tools.symlink(os.path.join(expdir, 'database.cfg'),
os.path.join(subexpdir, 'database.cfg'))
#if thresholdsarepresent:
# tools.symlink(os.path.join(expdir, 'word_thresholds.pkl'),
# os.path.join(subexpdir, 'word_thresholds.pkl'))
if not os.path.exists(os.path.join(subexpdir, 'trainfeats')):
trainutts = [names[i] for i in trainids[b][e]]
print 'number of examples: %d' % len(trainutts)
testutts = [names[i] for i in testids[b][e]]
#create the train and test sets
tools.writefile(os.path.join(subexpdir, 'trainfeats'),
{utt: features[utt]
for utt in trainutts})
tools.writefile(os.path.join(subexpdir, 'traintasks'),
{utt: taskstrings[utt]
for utt in trainutts})
tools.writefile(os.path.join(subexpdir, 'testfeats'),
{utt: features[utt]
for utt in testutts})
tools.writefile(os.path.join(subexpdir, 'testtasks'),
{utt: taskstrings[utt]
for utt in testutts})
if computing in ('condor', 'condor_gpu'):
#create the outputs directory
if not os.path.isdir(os.path.join(subexpdir, 'outputs')):
os.makedirs(os.path.join(subexpdir, 'outputs'))
if computing == 'condor_gpu':
jobfile = 'run_script_GPU.job'
else:
jobfile = 'run_script.job'
#only submit the job if it not running yet
in_queue = os.popen('if condor_q -nobatch -wide | grep -q %s; '
'then echo true; else echo false; fi' %
subexpdir).read().strip() == 'true'
#submit the condor job
if not in_queue:
os.system('condor_submit expdir=%s script=train_test'
' assist/condor/%s' % (subexpdir, jobfile))
else:
train_test.main(subexpdir)
newb = int(
np.floor((b + 1) * float(expconf['scale']) +
int(expconf['increment']) - 1))
newb = min(newb, len(blocks) - 2)
if b == newb:
break
else:
b = newb