def _compile_state_cpfs(
self, scope: Dict[str, TensorFluent], **kwargs
) -> List[CPFPair]:
"""Compiles the next state fluent CPFs given the current `state` and `action` scope.
Args:
scope (Dict[str, :obj:`rddl2tf.core.fluent.TensorFluent`]): The fluent scope for CPF evaluation.
batch_size (Optional[int]): The batch size.
Returns:
A list of state fluent CPFs compiled to :obj:`rddl2tf.core.fluent.TensorFluent`.
"""
noise = kwargs["noise"]
next_state_fluents = []
with self.graph.as_default():
with tf.name_scope("state_cpfs"):
for cpf in self.rddl.domain.state_cpfs:
cpf_noise = noise.get(cpf.name)
name_scope = utils.identifier(cpf.name)
with tf.name_scope(name_scope):
t = self._compile_expression(cpf.expr, scope, noise=cpf_noise)
next_state_fluents.append((cpf.name, t))
key = lambda f: self.rddl.domain.next_state_fluent_ordering.index(f[0])
next_state_fluents = sorted(next_state_fluents, key=key)
return next_state_fluents
def _compile_intermediate_cpfs(
self, scope: Dict[str, TensorFluent], **kwargs
) -> List[CPFPair]:
"""Compiles the intermediate fluent CPFs given the current `state` and `action` scope.
Args:
scope (Dict[str, :obj:`rddl2tf.core.fluent.TensorFluent`]): The fluent scope for CPF evaluation.
batch_size (Optional[int]): The batch size.
Returns:
A list of intermediate fluent CPFs compiled to :obj:`rddl2tf.core.fluent.TensorFluent`.
"""
noise = kwargs["noise"]
interm_fluents = []
with self.graph.as_default():
with tf.name_scope("intermediate_cpfs"):
for cpf in self.rddl.domain.intermediate_cpfs:
cpf_noise = noise.get(cpf.name)
name_scope = utils.identifier(cpf.name)
with tf.name_scope(name_scope):
t = self._compile_expression(cpf.expr, scope, noise=cpf_noise)
interm_fluents.append((cpf.name, t))
scope[cpf.name] = t
return interm_fluents
def _compile_pvariables(self,
pvariables: List[Tuple[str, str, np.array]]) -> List[Tuple[str, TensorFluent]]:
fluents = []
with self.graph.as_default():
for name, pvar_range, fluent in pvariables:
dtype = utils.range_type_to_dtype(pvar_range)
t = tf.constant(fluent, dtype=dtype, name=utils.identifier(name))
scope = [None] * len(t.shape)
fluent = TensorFluent(t, scope, batch=False)
fluents.append((name, fluent))
return fluents
def _compile_batch_fluents(self,
fluents: List[Tuple[str, TensorFluent]]) -> Sequence[tf.Tensor]:
'''Compiles `fluents` into tensors with given `batch_size`.
Returns:
Sequence[tf.Tensor]: A tuple of tensors with first dimension
corresponding to the batch size.
'''
batch_fluents = []
with self.graph.as_default():
for name, fluent in fluents:
name_scope = utils.identifier(name)
with tf.compat.v1.name_scope(name_scope):
t = tf.stack([fluent.tensor] * self.batch_size)
batch_fluents.append(t)
return tuple(batch_fluents)