def __init__(self, conf, output_dims, constraint, name=None):
'''EDDecoder constructor
Args:
conf: the decoder configuration as a ConfigParser
output_dims: a dictionary containing the output dimensions for each
output
constraint: the constraint for the variables
'''
#save the parameters
self.conf = dict(conf.items('decoder'))
#apply default configuration
default = os.path.join(os.path.dirname(os.path.realpath(__file__)),
'defaults',
type(self).__name__.lower() + '.cfg')
apply_defaults(self.conf, default)
self.outputs = output_dims.keys()
self.output_dims = output_dims
self.scope = tf.VariableScope(tf.AUTO_REUSE,
name or type(self).__name__,
constraint=constraint)
def __init__(self, conf):
'''Processor constructor
Args:
conf: processor configuration as a configparser
'''
self.conf = dict(conf.items('processor'))
#apply default configuration
default = os.path.join(os.path.dirname(os.path.realpath(__file__)),
'defaults',
type(self).__name__.lower() + '.cfg')
apply_defaults(self.conf, default)
def __init__(self,
conf,
dataconf,
modelconf,
evaluatorconf,
expdir,
server,
task_index):
'''
NnetTrainer constructor, creates the training graph
Args:
conf: the trainer config as a ConfigParser
dataconf: the data configuration as a ConfigParser
modelconf: the neural net model configuration
evaluatorconf: the evaluator configuration for evaluating
if None no evaluation will be done
expdir: directory where the summaries will be written
server: optional server to be used for distributed training
task_index: optional index of the worker task in the cluster
'''
#save some inputs
self.conf = dict(conf.items('trainer'))
#apply default configuration
default = os.path.join(
os.path.dirname(os.path.realpath(__file__)),
'defaults',
type(self).__name__.lower() + '.cfg')
apply_defaults(self.conf, default)
self.dataconf = dataconf
self.evaluatorconf = evaluatorconf
self.expdir = expdir
self.server = server
self.task_index = task_index
if ('norm_constraint' in self.conf
and self.conf['norm_constraint'] != 'None'):
constraint = constraints.MaxNorm(int(self.conf['norm_constraint']))
else:
constraint = None
#create the model
self.model = Model(
conf=modelconf,
trainlabels=int(self.conf['trainlabels']),
constraint=constraint)
def __init__(self, conf, model):
'''
Decoder constructor
Args:
conf: the decoder config
model: the model that will be used for decoding
'''
self.conf = dict(conf.items('decoder'))
#apply default configuration
default = os.path.join(os.path.dirname(os.path.realpath(__file__)),
'defaults',
type(self).__name__.lower() + '.cfg')
apply_defaults(self.conf, default)
self.model = model
def __init__(self, conf, constraint, name=None):
'''EDEncoder constructor
Args:
conf: the encoder configuration
name: the encoder name
constraint: the constraint for the variables
'''
#save the configuration
self.conf = dict(conf.items('encoder'))
#apply default configuration
default = os.path.join(os.path.dirname(os.path.realpath(__file__)),
'defaults',
type(self).__name__.lower() + '.cfg')
apply_defaults(self.conf, default)
self.scope = tf.VariableScope(tf.AUTO_REUSE,
name or type(self).__name__,
constraint=constraint)
def __init__(self, conf, dataconf, model):
'''Evaluator constructor
Args:
conf: the evaluator configuration as a ConfigParser
dataconf: the database configuration
model: the model to be evaluated
'''
self.conf = dict(conf.items('evaluator'))
#apply default configuration
default = os.path.join(os.path.dirname(os.path.realpath(__file__)),
'defaults',
type(self).__name__.lower() + '.cfg')
apply_defaults(self.conf, default)
self.model = model
#get the database configurations
inputs = self.model.input_names
input_sections = [self.conf[i].split(' ') for i in inputs]
self.input_dataconfs = []
for sectionset in input_sections:
self.input_dataconfs.append([])
for section in sectionset:
self.input_dataconfs[-1].append(dict(dataconf.items(section)))
targets = self.conf['targets'].split(' ')
if targets == ['']:
targets = []
target_sections = [self.conf[o].split(' ') for o in targets]
self.target_dataconfs = []
for sectionset in target_sections:
self.target_dataconfs.append([])
for section in sectionset:
self.target_dataconfs[-1].append(dict(dataconf.items(section)))