def test_mutate_input():
x = sym.Variable('x')
y = sym.conv2d(data=x, name='conv')
z = sym.assign(x, y)
t = sym.add(z, x)
try:
z = sym.assign(z, z)
assert False
except NNVMError:
pass
def test_mutate_input():
x = sym.Variable('x')
y = sym.conv2d(data=x, name='conv')
z = sym.assign(x, y)
t = sym.add(z, x)
try:
z = sym.assign(z, z)
assert False
except NNVMError:
pass
def minimize(self, obj):
variables = obj.list_input_variables()
grads = _base.gradients(obj, variables)
updates = []
for v, g in zip(variables, grads):
updates.append(_sym.assign(v, v + (-self.learning_rate) * g))
return _base.group(*updates)
def Variable(init, name=None):
if not isinstance(init, symbol.Symbol):
raise TypeError("Expect initialization expression to be Symbol")
name = NameManager.current.get(name, 'variable')
v = symbol.Variable(name)
_all_variable_inits.append(symbol.assign(v, init))
return v
def minimize(self, obj):
variables = obj.list_input_variables()
grads = _base.gradients(obj, variables)
updates = []
for v, g in zip(variables, grads):
updates.append(_sym.assign(v, v + (-self.learning_rate) * g))
return _base.group(*updates)
def Variable(init=None, name=None):
name = NameManager.current.get(name, 'variable')
v = symbol.Variable(name)
if init is not None:
if not isinstance(init, symbol.Symbol):
raise TypeError("Expect initialization expression to be Symbol")
_all_variable_inits.append(symbol.assign(v, init))
return v
def minimize(self, obj):
variables = obj.list_input_variables()
grads = _base.gradients(obj, variables)
updates = []
for i, v in enumerate(variables):
self.m.append(_base.Variable(_sym.zeros_like(v), self.name + '_m' + str(i)))
self.v.append(_base.Variable(_sym.zeros_like(v), self.name + '_v' + str(i)))
update_t = _sym.assign(self.t, self.t + 1)
rate = _sym.sqrt(1 - self.beta2 ** update_t) / (1 - self.beta1 ** update_t)
lr_t = self.learning_rate * rate
for var, g, m, v in zip(variables, grads, self.m, self.v):
update_m = _sym.assign(m, self.beta1 * m + (1 - self.beta1) * g)
update_v = _sym.assign(v, self.beta2 * v + (1 - self.beta2) * g * g)
update_var = _sym.assign(var,
var - lr_t * update_m / (_sym.sqrt(update_v) + self.epsilon))
updates.append(update_var)
return _base.group(*updates)
def test_list_args():
x = sym.Variable('x')
z = sym.Variable('z')
y = sym.conv2d(data=x, name='conv', dev='gpu')
y = sym.add(y, z, name='add1')
# write after read
z = sym.assign(x, y, name='assign')
assert z.list_inputs('read_only') == ['conv_weight', 'z']
assert z.list_inputs('aux_state') == ['x']
def test_list_args():
x = sym.Variable('x')
z = sym.Variable('z')
y = sym.conv2d(data=x, name='conv', dev='gpu')
y = sym.add(y, z, name='add1')
# write after read
z = sym.assign(x, y, name='assign')
assert z.list_input_names('read_only') == ['conv_weight', 'z']
assert z.list_input_names('aux_state') == ['x']
def minimize(self, obj):
variables = obj.list_input_variables()
grads = _base.gradients(obj, variables)
updates = []
for i, v in enumerate(variables):
self.m.append(
_base.Variable(_sym.zeros_like(v), self.name + '_m' + str(i)))
self.v.append(
_base.Variable(_sym.zeros_like(v), self.name + '_v' + str(i)))
update_t = _sym.assign(self.t, self.t + 1)
rate = _sym.sqrt(1 - self.beta2**update_t) / (1 - self.beta1**update_t)
lr_t = self.learning_rate * rate
for var, g, m, v in zip(variables, grads, self.m, self.v):
update_m = _sym.assign(m, self.beta1 * m + (1 - self.beta1) * g)
update_v = _sym.assign(v,
self.beta2 * v + (1 - self.beta2) * g * g)
update_var = _sym.assign(
var,
var - lr_t * update_m / (_sym.sqrt(update_v) + self.epsilon))
updates.append(update_var)
return _base.group(*updates)
def test_order_mutation_pass():
x = sym.Variable('x')
y = sym.conv2d(data=x, name='conv', dev='gpu')
y = sym.add(y, x, name='add1')
# write after read
z = sym.assign(x, y, name='assign')
# read after write
t = sym.add(y, x, name='add2')
g = graph.create(sym.Group([t, z]))
jgraph = json.loads(g.apply(['OrderMutation', 'SaveJSON']).json_attr('json'))
jnodes = jgraph['nodes']
nindex = {n['name']: i for i, n in enumerate(jnodes)}
assert nindex['assign'] in jnodes[nindex['add2']]['control_deps']
assert nindex['conv'] in jnodes[nindex['assign']]['control_deps']
assert nindex['add1'] in jnodes[nindex['assign']]['control_deps']
assert jnodes[nindex['assign']]['inputs'][0][2] == 1
def test_order_mutation_pass():
x = sym.Variable('x')
y = sym.conv2d(data=x, name='conv', dev='gpu')
y = sym.add(y, x, name='add1')
# write after read
z = sym.assign(x, y, name='assign')
# read after write
t = sym.add(y, x, name='add2')
g = graph.create(sym.Group([t, z]))
jgraph = json.loads(g.apply(['OrderMutation', 'SaveJSON']).json_attr('json'))
jnodes = jgraph['nodes']
nindex = {n['name']: i for i, n in enumerate(jnodes)}
assert nindex['assign'] in jnodes[nindex['add2']]['control_deps']
assert nindex['conv'] in jnodes[nindex['assign']]['control_deps']
assert nindex['add1'] in jnodes[nindex['assign']]['control_deps']
assert jnodes[nindex['assign']]['inputs'][0][2] == 1
def test_control_dep():
x = sym.Variable('x')
y = sym.conv2d(data=x, name='conv')
z = sym.assign(x, y)
t = sym.add(x, x)
t._add_control_deps([z, y])
def test_control_dep():
x = sym.Variable('x')
y = sym.conv2d(data=x, name='conv')
z = sym.assign(x, y)
t = sym.add(x, x)
t._add_control_deps([z, y])