suppress errors in beanmachine/beanmachine/ppl

Differential Revision: D33201580

fbshipit-source-id: 139d7ba56f8488da5192f44ce4e3f49262e4cb61

src/beanmachine/ppl/experimental/inference_compilation/ic_infer.py

@@ -117,6 +117,7 @@ def do_adaptation(
         loss = -(proposal_distribution.log_prob(node_var.value))
         optimizer.zero_grad()
         loss.backward()
+        # pyre-fixme[20]: Argument `closure` expected.
         optimizer.step()
 
 
@@ -419,7 +420,6 @@ def _proposer_func(
                     raise Exception("No observation embedding network found!")
 
                 obs_vec = torch.stack(obs_nodes, dim=0).flatten()
-                # pyre-fixme
                 obs_embedding = obs_embedding_net.forward(obs_vec)
 
             node_embedding_nets = self._node_embedding_nets
@@ -429,7 +429,6 @@ def _proposer_func(
             mb_embedding = torch.zeros(self._MB_EMBEDDING_DIM)
             mb_nodes = list(
                 map(
-                    # pyre-fixme[29]: `Union[Tensor, nn.Module]` is not a function.
                     lambda mb_node: node_embedding_nets(mb_node).forward(
                         utils.ensure_1d(
                             world.get_node_in_world_raise_error(mb_node).value
@@ -449,15 +448,11 @@ def _proposer_func(
                 mb_vec = torch.stack(mb_nodes, dim=0).unsqueeze(1)
                 # TODO: try pooling rather than just slicing out last hidden
                 mb_embedding = utils.ensure_1d(
-                    # pyre-fixme[29]: `Union[Tensor, nn.Module]` is not a function.
-                    mb_embedding_nets(node)
-                    .forward(mb_vec)[0][-1, :, :]
-                    .squeeze()
+                    mb_embedding_nets(node).forward(mb_vec)[0][-1, :, :].squeeze()
                 )
             node_proposal_param_nets = self._node_proposal_param_nets
             if node_proposal_param_nets is None:
                 raise Exception("No node proposal parameter networks found!")
-            # pyre-fixme[29]: `Union[Tensor, nn.Module]` is not a function.
             param_vec = node_proposal_param_nets(node).forward(
                 torch.cat((mb_embedding, obs_embedding))
             )
@@ -475,9 +470,7 @@ def _proposal_distribution_for_node(
         """
         node_var = self.world_.get_node_in_world_raise_error(node)
         distribution = node_var.distribution
-        # pyre-fixme
         sample_val = distribution.sample()
-        # pyre-fixme
         support = distribution.support
 
         ndim = sample_val.dim()

src/beanmachine/ppl/experimental/neutra/iaflayer.py

@@ -84,6 +84,10 @@ def __init__(
         self.loga_max_clip_ = loga_max_clip
         self.stable_ = stable
 
+    # pyre-fixme[14]: `get_parameter` overrides method defined in `Module`
+    #  inconsistently.
+    # pyre-fixme[15]: `get_parameter` overrides method defined in `Module`
+    #  inconsistently.
     def get_parameter(self, x: Tensor) -> Tuple[Tensor, Tensor]:
 
         """

src/beanmachine/ppl/experimental/neutra/maskedlinear.py

@@ -67,6 +67,7 @@ def set_mask(self, mask: Tensor) -> None:
             raise ValueError("Dimension mismatches between mask and layer.")
         self.mask.data.copy_(mask.t())
 
+    # pyre-fixme[14]: `forward` overrides method defined in `Linear` inconsistently.
     def forward(self, input_: Tensor) -> Tensor:
         """
         the forward method that does the masked linear computation

src/beanmachine/ppl/experimental/vi/mean_field_variational_approximation.py

@@ -63,7 +63,7 @@ def __init__(
         if not base_args:
             base_args = {}
 
-        event_shape = target_dist.event_shape  # pyre-ignore[16]
+        event_shape = target_dist.event_shape
         # form independent product distribution of `base_dist` for `event_shape`
         if len(event_shape) == 0:
             self.base_args = base_args
@@ -105,14 +105,14 @@ def _base_dist(**kwargs):
         )
 
         # unwrap nested independents before setting transform
-        support = target_dist.support  # pyre-ignore[16]
+        support = target_dist.support
         while isinstance(support, constraints.independent):
             support = support.base_constraint
         self._transform = biject_to(support)
 
         super().__init__(
-            self.new_dist.batch_shape,  # pyre-ignore
-            self.new_dist.event_shape,  # pyre-ignore
+            self.new_dist.batch_shape,
+            self.new_dist.event_shape,
             validate_args=validate_args,
         )
 

src/beanmachine/ppl/inference/compositional_infer.py

@@ -59,7 +59,6 @@ def get_proposers(
         proposers = []
         for node in target_rvs:
             if node not in self._proposers:
-                # pyre-ignore[16]
                 support = world.get_variable(node).distribution.support
                 if any(
                     is_constraint_eq(

src/beanmachine/ppl/inference/proposer/base_single_site_proposer.py

@@ -32,7 +32,6 @@ def propose(self, world: World):
         """
         proposal_dist = forward_dist = self.get_proposal_distribution(world)
         old_value = world[self.node]
-        # pyre-ignore[20]
         proposed_value = proposal_dist.sample()
         new_world = world.replace({self.node: proposed_value})
         backward_dist = self.get_proposal_distribution(new_world)

src/beanmachine/ppl/inference/proposer/single_site_random_walk_proposer.py

@@ -44,7 +44,7 @@ def do_adaptation(self, world, accept_log_prob, *args, **kwargs) -> None:
     def get_proposal_distribution(self, world: World) -> dist.Distribution:
         """Propose a new value for self.node using the prior distribution."""
         node = world.get_variable(self.node)
-        node_support = node.distribution.support  # pyre-ignore [16]
+        node_support = node.distribution.support
 
         if is_constraint_eq(node_support, dist.constraints.real):
             return dist.Normal(node.value, self.step_size)

src/beanmachine/ppl/inference/single_site_nmc.py

@@ -70,7 +70,7 @@ def _init_nmc_proposer(self, node: RVIdentifier, world: World) -> BaseProposer:
         of NMC proposer will be chosen based on a node's support.
         """
         distribution = world.get_variable(node).distribution
-        support = distribution.support  # pyre-ignore
+        support = distribution.support
         if is_constraint_eq(support, dist.constraints.real):
             return SingleSiteRealSpaceNMCProposer(node, self.alpha, self.beta)
         elif is_constraint_eq(support, dist.constraints.greater_than):

src/beanmachine/ppl/legacy/inference/proposer/single_site_newtonian_monte_carlo_proposer.py

@@ -89,7 +89,6 @@ def get_proposal_distribution(
         that was used or needs to be used to find the proposal distribution
         """
         if node not in self.proposers_:
-            # pyre-fixme
             node_distribution_support = node_var.distribution.support
             if world.get_transforms_for_node(
                 node

src/beanmachine/ppl/legacy/inference/proposer/single_site_random_walk_proposer.py

@@ -108,7 +108,6 @@ def get_proposal_distribution(
         if world.get_transforms_for_node(
             node
         ).transform_type != TransformType.NONE or is_constraint_eq(
-            # pyre-fixme
             node_distribution.support,
             dist.constraints.real,
         ):

src/beanmachine/ppl/legacy/inference/proposer/single_site_uniform_proposer.py

@@ -46,7 +46,6 @@ def get_proposal_distribution(
         node_distribution = node_var.distribution
         if (
             is_constraint_eq(
-                # pyre-fixme
                 node_distribution.support,
                 dist.constraints.boolean,
             )

src/beanmachine/ppl/legacy/world/variable.py

@@ -94,7 +94,6 @@ def bar(self):
 
     @property
     def is_discrete(self) -> bool:
-        # pyre-fixme
         return self.distribution.support.is_discrete
 
     def __post_init__(self) -> None:

src/beanmachine/ppl/world/utils.py

@@ -101,7 +101,6 @@ def get_default_transforms(distribution: Distribution) -> dist.Transform:
     :returns: a Transform that need to be applied to the distribution
         to transform it from constrained space into unconstrained space
     """
-    # pyre-fixme
     if distribution.support.is_discrete:
         return dist.transforms.identity_transform
     else:
@@ -115,11 +114,9 @@ def initialize_value(distribution: Distribution, initialize_from_prior: bool = F
     :param initialize_from_prior: if true, returns sample from prior
     :returns: the value to the set the Variable value to
     """
-    # pyre-fixme
     sample_val = distribution.sample()
     if initialize_from_prior:
         return sample_val
-    # pyre-fixme
     support = distribution.support
     if isinstance(support, dist.constraints.independent):
         support = support.base_constraint

src/beanmachine/ppl/world/world.py

@@ -206,10 +206,8 @@ def enumerate_node(self, node: RVIdentifier) -> torch.Tensor:
           A tensor enumerating the support of the node.
         """
         distribution = self._variables[node].distribution
-        # pyre-ignore[16]
         if not distribution.has_enumerate_support:
             raise ValueError(str(node) + " is not enumerable")
-        # pyre-ignore[16]
         return distribution.enumerate_support()
 
     def _run_node(