def __init__(self,
name,
learnable,
ranges,
is_observed=False,
**kwargs): #TODO: code duplication here
self.name = name
self._evaluated = False
self._observed = is_observed
self._observed_value = None
self._current_value = None
self.construct_deterministic_parents(learnable, ranges, kwargs)
self.parents = join_sets_list(
[var2link(x).vars for x in kwargs.values()])
self.ancestors = join_sets_list(
[self.parents] + [parent.ancestors for parent in self.parents])
self.link = LinkConstructor(**kwargs)
self.samples = None
self.ranges = {}
self.dataset = None
self.has_random_dataset = False
self.has_observed_value = False
self.is_normalized = True
self.partial_links = {
name: var2link(link)
for name, link in kwargs.items()
}
def __init__(self, name, learnable, ranges, is_observed=False, has_bias=False, is_policy=False, is_reward=False, **kwargs):
self._input = kwargs
self.name = name
self._evaluated = False
self._observed = is_observed
self._observed_value = None
self._current_value = None
#if self._check_for_stochastic_process_arguments():
# return
if has_bias:
self.construct_biases(learnable, ranges, kwargs)
self.construct_deterministic_parents(learnable, ranges, kwargs)
self.parents = join_sets_list([var2link(x).vars for x in kwargs.values()])
self.link = Link(**kwargs)
self.ancestors = join_sets_list([self.parents] + [parent.ancestors for parent in self.parents])
self.samples = None
self.ranges = {}
self.dataset = None
self.has_random_dataset = False
self.has_observed_value = False
self.is_normalized = True
self.silenced = False
self.partial_links = {name: var2link(link) for name, link in kwargs.items()}
# Reinforcement learning descriptors
if not (is_policy and is_reward):
self._policy = is_policy
self._reward = is_reward
else:
raise ValueError("A variable cannot be both a RL policy and a RL reward")
def __init__(self,
name,
learnable,
ranges,
is_observed=False,
has_bias=False,
**kwargs):
self._input = kwargs.copy()
self.name = name
self.learnable = learnable
self._bias = has_bias
self._evaluated = False
self._observed = is_observed
self._observed_value = None
self._current_value = None
if self._check_for_stochastic_process_arguments():
return
if has_bias:
self.construct_biases(learnable, ranges, kwargs)
self.construct_deterministic_parents(learnable, ranges, kwargs)
self.parents = join_sets_list(
[var2link(x).vars for x in kwargs.values()])
self.link = Link(**kwargs)
self.ancestors = join_sets_list(
[self.parents] + [parent.ancestors for parent in self.parents])
self.samples = None
self.ranges = {}
self.dataset = None
self.has_random_dataset = False
self.has_observed_value = False
self.is_normalized = True
self.partial_links = {
name: var2link(link)
for name, link in kwargs.items()
}
def __call__(self, *args, **kwargs):
link_args = [var2link(arg) for arg in args]
link_kwargs = {name: var2link(arg) for name, arg in kwargs.items()}
arg_vars = {
var
for link in link_args if isinstance(link, PartialLink)
for var in link.vars
}
kwarg_vars = {
var
for _, link in link_kwargs.items()
if isinstance(link, PartialLink) for var in link.vars
}
def fn(values):
args = [
x.fn(values) if isinstance(x, PartialLink) else x
for x in link_args
]
kwargs = dict({(name,
x.fn(values)) if isinstance(x, PartialLink) else
(name, x)
for name, x in link_kwargs.items()})
return self.fn(*args, **kwargs)
return PartialLink(arg_vars.union(kwarg_vars), fn, self.links)
def __init__(self, value, name, log_determinant=None, learnable=False, has_bias=False, is_observed=False, variable_range=geometric_ranges.UnboundedRange(),
is_policy=False, is_reward=False):
self._type = "Deterministic node"
if not isinstance(log_determinant, PartialLink):
if log_determinant is None:
log_determinant = torch.tensor(np.zeros((1, 1))).float().to(device)
var2link(log_determinant)
ranges = {"value": variable_range,
"log_determinant": geometric_ranges.UnboundedRange()}
super().__init__(name, value=value, log_determinant=log_determinant, learnable=learnable, has_bias=has_bias,
ranges=ranges, is_observed=is_observed, is_policy=is_policy, is_reward=is_reward)
self.distribution = distributions.DeterministicDistribution()
def __init__(self):
self.kwargs = kwargs
links = [
link for partial_link in kwargs.values()
for link in var2link(partial_link).links
]
super().__init__(*links)
def __init__(self, name, learnable, ranges, is_observed=False, **kwargs):
class VarLink(chainer.ChainList):
def __init__(self):
self.kwargs = kwargs
links = [
link for partial_link in kwargs.values()
for link in var2link(partial_link).links
]
super().__init__(*links)
def __call__(self, values):
return {
k: var2link(x).fn(values)
for k, x in self.kwargs.items()
}
self.name = name
self._evaluated = False
self._observed = is_observed
self._observed_value = None
self._current_value = None
self.construct_deterministic_parents(learnable, ranges, kwargs)
self.parents = join_sets_list(
[var2link(x).vars for x in kwargs.values()])
self.link = VarLink()
self.samples = []
self.ranges = {}
self.dataset = None
self.has_random_dataset = False
self.has_observed_value = False
def __init__(self, **kwargs):
self.kwargs = kwargs
modules = [
link for partial_link in kwargs.values()
for link in var2link(partial_link).links
]
super().__init__(
modules
) #TODO: asserts that specified links are valid pytorch modules
def __init__(self, function, links, *args, **kwargs):
self.function = function
self.link_args = [var2link(arg) for arg in args]
self.link_kwargs = {
name: var2link(arg)
for name, arg in kwargs.items()
}
arg_vars = {
var
for link in self.link_args if isinstance(link, PartialLink)
for var in link.vars
}
kwarg_vars = {
var
for _, link in self.link_kwargs.items()
if isinstance(link, PartialLink) for var in link.vars
}
super().__init__(vars=arg_vars.union(kwarg_vars), links=links)
def __call__(self, values):
return {k: var2link(x).fn(values) for k, x in self.kwargs.items()}
def _get_string(self, *args, **kwargs):
return self.name + "(" + ", ".join([
var2link(a).__str__()
for n, a in enumerate(list(args) + list(kwargs.values()))
]) + ")"
def __str__(self):
return str(self.function) + "(" + ", ".join([
str(var2link(a)) for n, a in enumerate(
list(self.link_args) + list(self.link_kwargs.values()))
]) + ")"
