Python theano.gof.local_optimizer() Examples
The following are 12
code examples of theano.gof.local_optimizer().
You can vote up the ones you like or vote down the ones you don't like,
and go to the original project or source file by following the links above each example.
You may also want to check out all available functions/classes of the module
theano.gof
, or try the search function
.
.
Example #1
| Source File: test_debugmode.py From D-VAE with MIT License | 6 votes |
|
def test_badoptimization():
@gof.local_optimizer([theano.tensor.add])
def insert_broken_add(node):
if node.op == theano.tensor.add:
return [off_by_half(*node.inputs)]
return False
edb = gof.EquilibriumDB()
edb.register('insert_broken_add', insert_broken_add, 'all')
opt = edb.query('+all')
a = theano.tensor.dvector()
b = theano.tensor.dvector()
f = theano.function([a, b], a + b,
mode=debugmode.DebugMode(optimizer=opt))
try:
f([1.0, 2.0, 3.0], [2, 3, 4],)
except debugmode.BadOptimization as e:
assert str(e.reason) == 'insert_broken_add'
return # TEST PASS
assert False
Example #2
| Source File: test_debugmode.py From attention-lvcsr with MIT License | 6 votes |
|
def test_badoptimization():
@gof.local_optimizer([theano.tensor.add])
def insert_broken_add(node):
if node.op == theano.tensor.add:
return [off_by_half(*node.inputs)]
return False
edb = gof.EquilibriumDB()
edb.register('insert_broken_add', insert_broken_add, 'all')
opt = edb.query('+all')
a = theano.tensor.dvector()
b = theano.tensor.dvector()
f = theano.function([a, b], a + b,
mode=debugmode.DebugMode(optimizer=opt))
try:
f([1.0, 2.0, 3.0], [2, 3, 4],)
except debugmode.BadOptimization as e:
assert str(e.reason) == 'insert_broken_add'
return # TEST PASS
assert False
Example #3
| Source File: test_debugmode.py From D-VAE with MIT License | 5 votes |
|
def test_badoptimization_opt_err():
"""This variant of test_badoptimization() replace the working code
with a new apply node that will raise an error.
"""
@gof.local_optimizer([theano.tensor.add])
def insert_bigger_b_add(node):
if node.op == theano.tensor.add:
inputs = list(node.inputs)
if inputs[-1].owner is None:
inputs[-1] = theano.tensor.concatenate((inputs[-1],
inputs[-1]))
return [node.op(*inputs)]
return False
edb = gof.EquilibriumDB()
edb.register('insert_bigger_b_add', insert_bigger_b_add, 'all')
opt = edb.query('+all')
a = theano.tensor.dvector()
b = theano.tensor.dvector()
f = theano.function([a, b], a + b,
mode=debugmode.DebugMode(optimizer=opt))
try:
f([1.0, 2.0, 3.0], [2, 3, 4],)
except Exception as e:
assert 'insert_bigger_b_add' in exc_message(e)
return # TEST PASS
assert False
Example #4
| Source File: test_debugmode.py From D-VAE with MIT License | 5 votes |
|
def test_stochasticoptimization():
# this optimization alternates between triggering and not triggering.
last_time_replaced = [False]
@gof.local_optimizer([theano.tensor.add])
def insert_broken_add_sometimes(node):
if node.op == theano.tensor.add:
last_time_replaced[0] = not last_time_replaced[0]
if last_time_replaced[0]:
return [off_by_half(*node.inputs)]
return False
edb = gof.EquilibriumDB()
edb.register(
'insert_broken_add_sometimes',
insert_broken_add_sometimes,
'all')
opt = edb.query('+all')
a = theano.tensor.dvector()
b = theano.tensor.dvector()
try:
theano.function([a, b],
theano.tensor.add(a, b),
mode=debugmode.DebugMode(
optimizer=opt,
check_c_code=True,
stability_patience=max(2, config.DebugMode.patience)))
except debugmode.StochasticOrder:
return # TEST PASS
assert False
Example #5
| Source File: opt_util.py From D-VAE with MIT License | 5 votes |
|
def alpha_merge(cls, alpha_in, beta_in):
def wrapper(maker):
@local_optimizer([GpuElemwise])
@wraps(maker)
def opt(node):
if (isinstance(node.op, GpuElemwise) and
node.op.scalar_op == scal.mul and
node.nin == 2):
targ = find_node(node.inputs[0], cls)
if targ is None:
targ = find_node(node.inputs[1], cls)
if targ is None:
return
lr = grab_cpu_scalar(node.inputs[0],
nd=targ.outputs[0].ndim)
else:
lr = grab_cpu_scalar(node.inputs[1],
nd=targ.outputs[0].ndim)
if lr is None or targ is None:
return None
inputs = list(targ.inputs)
try:
c = get_scalar_constant_value(lr)
if c == 0:
inputs[alpha_in] = lr
inputs[beta_in] = lr
elif c == 1:
inputs[alpha_in] = targ.inputs[alpha_in]
inputs[beta_in] = targ.inputs[beta_in]
else:
inputs[alpha_in] = lr * targ.inputs[alpha_in]
inputs[beta_in] = lr * targ.inputs[beta_in]
except NotScalarConstantError:
inputs[alpha_in] = lr * targ.inputs[alpha_in]
inputs[beta_in] = lr * targ.inputs[beta_in]
return maker(targ, *inputs)
return opt
return wrapper
Example #6
| Source File: opt_util.py From D-VAE with MIT License | 5 votes |
|
def output_merge(cls, alpha_in, beta_in, out_in):
def wrapper(maker):
@local_optimizer([GpuElemwise])
@wraps(maker)
def opt(node):
if (isinstance(node.op, GpuElemwise) and
node.op.scalar_op == scal.add and
node.nin == 2):
targ = find_node(node.inputs[0], cls)
W = node.inputs[1]
if targ is None:
targ = find_node(node.inputs[1], cls)
W = node.inputs[0]
if targ is None:
return None
if not is_equal(targ.inputs[beta_in], 0.0):
# other cases are too complex for now
return None
if W.broadcastable != targ.inputs[out_in].broadcastable:
# May change later to do the broadcast, but it's
# under discussion.
return None
inputs = list(targ.inputs)
inputs[out_in] = W
inputs[beta_in] = _one.clone()
return maker(targ, *inputs)
return opt
return wrapper
Example #7
| Source File: test_debugmode.py From attention-lvcsr with MIT License | 5 votes |
|
def test_badoptimization_opt_err():
"""This variant of test_badoptimization() replace the working code
with a new apply node that will raise an error.
"""
@gof.local_optimizer([theano.tensor.add])
def insert_bigger_b_add(node):
if node.op == theano.tensor.add:
inputs = list(node.inputs)
if inputs[-1].owner is None:
inputs[-1] = theano.tensor.concatenate((inputs[-1],
inputs[-1]))
return [node.op(*inputs)]
return False
edb = gof.EquilibriumDB()
edb.register('insert_bigger_b_add', insert_bigger_b_add, 'all')
opt = edb.query('+all')
a = theano.tensor.dvector()
b = theano.tensor.dvector()
f = theano.function([a, b], a + b,
mode=debugmode.DebugMode(optimizer=opt))
try:
f([1.0, 2.0, 3.0], [2, 3, 4],)
except Exception as e:
assert 'insert_bigger_b_add' in exc_message(e)
return # TEST PASS
assert False
Example #8
| Source File: test_debugmode.py From attention-lvcsr with MIT License | 5 votes |
|
def test_stochasticoptimization():
# this optimization alternates between triggering and not triggering.
last_time_replaced = [False]
@gof.local_optimizer([theano.tensor.add])
def insert_broken_add_sometimes(node):
if node.op == theano.tensor.add:
last_time_replaced[0] = not last_time_replaced[0]
if last_time_replaced[0]:
return [off_by_half(*node.inputs)]
return False
edb = gof.EquilibriumDB()
edb.register(
'insert_broken_add_sometimes',
insert_broken_add_sometimes,
'all')
opt = edb.query('+all')
a = theano.tensor.dvector()
b = theano.tensor.dvector()
try:
theano.function([a, b],
theano.tensor.add(a, b),
mode=debugmode.DebugMode(
optimizer=opt,
check_c_code=True,
stability_patience=max(2, config.DebugMode.patience)))
except debugmode.StochasticOrder:
return # TEST PASS
assert False
Example #9
| Source File: opt_util.py From attention-lvcsr with MIT License | 5 votes |
|
def alpha_merge(cls, alpha_in, beta_in):
def wrapper(maker):
@local_optimizer([GpuElemwise])
@wraps(maker)
def opt(node):
if (isinstance(node.op, GpuElemwise) and
node.op.scalar_op == scal.mul and
node.nin == 2):
targ = find_node(node.inputs[0], cls)
if targ is None:
targ = find_node(node.inputs[1], cls)
if targ is None:
return
lr = grab_cpu_scalar(node.inputs[0],
nd=targ.outputs[0].ndim)
else:
lr = grab_cpu_scalar(node.inputs[1],
nd=targ.outputs[0].ndim)
if lr is None or targ is None:
return None
inputs = list(targ.inputs)
try:
c = get_scalar_constant_value(lr)
if c == 0:
inputs[alpha_in] = lr
inputs[beta_in] = lr
elif c == 1:
inputs[alpha_in] = targ.inputs[alpha_in]
inputs[beta_in] = targ.inputs[beta_in]
else:
inputs[alpha_in] = lr * targ.inputs[alpha_in]
inputs[beta_in] = lr * targ.inputs[beta_in]
except NotScalarConstantError:
inputs[alpha_in] = lr * targ.inputs[alpha_in]
inputs[beta_in] = lr * targ.inputs[beta_in]
return maker(targ, *inputs)
return opt
return wrapper
Example #10
| Source File: opt_util.py From attention-lvcsr with MIT License | 5 votes |
|
def output_merge(cls, alpha_in, beta_in, out_in):
def wrapper(maker):
@local_optimizer([GpuElemwise])
@wraps(maker)
def opt(node):
if (isinstance(node.op, GpuElemwise) and
node.op.scalar_op == scal.add and
node.nin == 2):
targ = find_node(node.inputs[0], cls)
W = node.inputs[1]
if targ is None:
targ = find_node(node.inputs[1], cls)
W = node.inputs[0]
if targ is None:
return None
if not is_equal(targ.inputs[beta_in], 0.0):
# other cases are too complex for now
return None
if W.broadcastable != targ.inputs[out_in].broadcastable:
# May change later to do the broadcast, but it's
# under discussion.
return None
inputs = list(targ.inputs)
inputs[out_in] = W
inputs[beta_in] = _one.clone()
return maker(targ, *inputs)
return opt
return wrapper
Example #11
| Source File: opt.py From D-VAE with MIT License | 4 votes |
|
def op_lifter(OP, cuda_only=False):
"""
OP(..., host_from_gpu(), ...) -> host_from_gpu(GpuOP(...))
gpu_from_host(OP(inp0, ...)) -> GpuOP(inp0, ...)
"""
def f(maker):
def local_opt(node):
if type(node.op) in OP:
# Either one of our inputs is on the gpu or
# all of our clients are on the gpu
replace = False
# TODO: Maybe set context_name with infer_context_name()?
context_name = None
# We replace if any input is a host_from_gpu
for i in node.inputs:
if i.owner and i.owner.op == host_from_gpu:
context_name = i.owner.inputs[0].type.context_name
replace = True
break
if not replace:
# We replace if *all* clients are on the GPU
clients = [c for o in node.outputs for c in o.clients]
replace = len(clients) != 0
for c, idx in clients:
if (c == 'output' or
not isinstance(c.op, GpuFromHost)):
replace = False
# TODO: check that the clients want the same context?
if replace:
# All clients are GpuFromHost and we have at least one
context_name = clients[0][0].op.context_name
# Check if we should replace
if (not replace or
(cuda_only and
get_context(context_name).kind != 'cuda')):
return False
# tag the inputs with the context in case
# the context was derived from the outputs
for i in node.inputs:
i.tag.context_name = context_name
new_op = maker(node, context_name)
# This is needed as sometimes new_op inherits from OP.
if new_op and new_op != node.op:
if isinstance(new_op, theano.Op):
return [safe_to_cpu(o) for o in
new_op(*node.inputs, return_list=True)]
elif isinstance(new_op, (tuple, list)):
return [safe_to_cpu(o) for o in new_op]
else: # suppose it is a variable on the GPU
return [host_from_gpu(new_op)]
return False
local_opt.__name__ = maker.__name__
return local_optimizer(OP)(local_opt)
return f
Example #12
| Source File: opt.py From attention-lvcsr with MIT License | 4 votes |
|
def op_lifter(OP, cuda_only=False):
"""
OP(..., host_from_gpu(), ...) -> host_from_gpu(GpuOP(...))
gpu_from_host(OP(inp0, ...)) -> GpuOP(inp0, ...)
"""
def f(maker):
def local_opt(node):
if type(node.op) in OP:
# Either one of our inputs is on the gpu or
# all of our clients are on the gpu
replace = False
# TODO: Maybe set context_name with infer_context_name()?
context_name = None
# We replace if any input is a host_from_gpu
for i in node.inputs:
if i.owner and i.owner.op == host_from_gpu:
context_name = i.owner.inputs[0].type.context_name
replace = True
break
if not replace:
# We replace if *all* clients are on the GPU
clients = [c for o in node.outputs for c in o.clients]
replace = len(clients) != 0
for c, idx in clients:
if (c == 'output' or
not isinstance(c.op, GpuFromHost)):
replace = False
# TODO: check that the clients want the same context?
if replace:
# All clients are GpuFromHost and we have at least one
context_name = clients[0][0].op.context_name
# Check if we should replace
if (not replace or
(cuda_only and
get_context(context_name).kind != 'cuda')):
return False
# tag the inputs with the context in case
# the context was derived from the outputs
for i in node.inputs:
i.tag.context_name = context_name
new_op = maker(node, context_name)
# This is needed as sometimes new_op inherits from OP.
if new_op and new_op != node.op:
if isinstance(new_op, theano.Op):
return [safe_to_cpu(o) for o in
new_op(*node.inputs, return_list=True)]
elif isinstance(new_op, (tuple, list)):
return [safe_to_cpu(o) for o in new_op]
else: # suppose it is a variable on the GPU
return [host_from_gpu(new_op)]
return False
local_opt.__name__ = maker.__name__
return local_optimizer(OP)(local_opt)
return f
