def wrap(f):
try:
if is_tf():
gdef = tf.Graph().as_default()
gdef.__enter__()
tf_reset_session()
executor = f()
ts = time.time()
for _ in range(n):
executor()
te = time.time()
print 'func:%r %d times took: %2.4f sec' % (f.__name__, n, te-ts)
if is_tf():
gdef.__exit__(None, None, None)
except Exception as e:
print e
def wrap(f):
try:
if is_tf():
gdef = tf.Graph().as_default()
gdef.__enter__()
tf_reset_session()
executor = f()
ts = time.time()
for _ in range(n):
executor()
te = time.time()
print 'func:%r %d times took: %2.4f sec' % (f.__name__, n, te - ts)
if is_tf():
gdef.__exit__(None, None, None)
except Exception as e:
print e
def is_variable(x):
if is_theano():
return isinstance(x, theano.gof.Variable)
elif is_cgt():
return isinstance(x, cgt.core.Node)
elif is_tf():
return isinstance(x, (tf.Tensor, tf.Variable))
else:
import ipdb; ipdb.set_trace()
def is_variable(x):
if is_theano():
return isinstance(x, theano.gof.Variable)
elif is_cgt():
return isinstance(x, cgt.core.Node)
elif is_tf():
return isinstance(x, (tf.Tensor, tf.Variable))
else:
import ipdb
ipdb.set_trace()
def get_inputs(outputs):
if is_theano():
return theano.gof.graph.inputs(outputs)
elif is_cgt():
outputs = list(outputs)
return [node for node in cgt.core.topsorted(outputs) if node.is_input()]
elif is_tf():
outputs = list(outputs)
return [node for node in _tf_topsorted(outputs) if _tf_is_input(node)]
else:
import ipdb; ipdb.set_trace()
def set_value(x, val):
"""
Get parameter value from a shared variable.
"""
if is_theano():
x.set_value(val)
elif is_cgt():
x.op.set_value(val)
elif is_tf():
tf.get_session().run(tf.assign(x, val))
else:
import ipdb; ipdb.set_trace()
def shape(x):
if is_theano():
return x.shape
elif is_cgt():
return x.shape
elif is_tf():
if isinstance(x, (tf.Tensor, tf.Variable)):
return x.shape
else:
import ipdb; ipdb.set_trace()
else:
import ipdb; ipdb.set_trace()
def set_value(x, val):
"""
Get parameter value from a shared variable.
"""
if is_theano():
x.set_value(val)
elif is_cgt():
x.op.set_value(val)
elif is_tf():
tf.get_session().run(tf.assign(x, val))
else:
import ipdb
ipdb.set_trace()
def shape(x):
if is_theano():
return x.shape
elif is_cgt():
return x.shape
elif is_tf():
if isinstance(x, (tf.Tensor, tf.Variable)):
return x.shape
else:
import ipdb
ipdb.set_trace()
else:
import ipdb
ipdb.set_trace()
def get_inputs(outputs):
if is_theano():
return theano.gof.graph.inputs(outputs)
elif is_cgt():
outputs = list(outputs)
return [
node for node in cgt.core.topsorted(outputs) if node.is_input()
]
elif is_tf():
outputs = list(outputs)
return [node for node in _tf_topsorted(outputs) if _tf_is_input(node)]
else:
import ipdb
ipdb.set_trace()
def shape(x):
if is_theano():
return x.shape
elif is_cgt():
return x.shape
elif is_tf():
if isinstance(x, (tf.Tensor, tf.Variable)):
return x.shape
else:
import ipdb; ipdb.set_trace()
else:
import ipdb; ipdb.set_trace()
if is_tf():
_tf_session = None
_tf_blank_vars = []
def tf_get_session():
global _tf_session
if _tf_session is None:
_tf_session = tf.Session()
return _tf_session
def tf_reset_session():
global _tf_session
if _tf_session:
_tf_session.close()
_tf_session = tf.Session()
from tensorfuse.config import is_theano, is_cgt, is_tf, is_mxnet
if is_theano():
from tensorfuse.backend.theano.tensor.signal import *
elif is_cgt():
from tensorfuse.backend.cgt.tensor.signal import *
elif is_tf():
from tensorfuse.backend.tensorflow.tensor.signal import *
elif is_mxnet():
from tensorfuse.backend.mxnet.tensor.signal import *
else:
raise ValueError('Unknown backend')