fix pyright errors

sbi/analysis/plot.py

@@ -19,7 +19,7 @@
 
 try:
     collectionsAbc = collections.abc  # type: ignore
-except:
+except AttributeError:
     collectionsAbc = collections
 
 
@@ -40,9 +40,9 @@ def _update(d, u):
     # https://stackoverflow.com/a/3233356
     for k, v in six.iteritems(u):
         dv = d.get(k, {})
-        if not isinstance(dv, collectionsAbc.Mapping):  # tpye: ignore
+        if not isinstance(dv, collectionsAbc.Mapping):  # type: ignore
             d[k] = v
-        elif isinstance(v, collectionsAbc.Mapping):  # tpye: ignore
+        elif isinstance(v, collectionsAbc.Mapping):  # type: ignore
             d[k] = _update(dv, v)
         else:
             d[k] = v

sbi/samplers/rejection/rejection.py

@@ -192,7 +192,7 @@ def accept_reject_sample(
     warn_acceptance: float = 0.01,
     sample_for_correction_factor: bool = False,
     max_sampling_batch_size: int = 10_000,
-    proposal_sampling_kwargs: Dict = {},
+    proposal_sampling_kwargs: Optional[Dict] = None,
     alternative_method: Optional[str] = None,
     **kwargs,
 ) -> Tuple[Tensor, Tensor]:
@@ -264,12 +264,12 @@ def accept_reject_sample(
         if isinstance(proposal, nn.Module):
             candidates = proposal.sample(
                 sampling_batch_size,
-                **proposal_sampling_kwargs,  # type: ignore
+                **proposal_sampling_kwargs or {},  # type: ignore
             ).reshape(sampling_batch_size, -1)
         else:
             candidates = proposal.sample(
-                (sampling_batch_size,),
-                **proposal_sampling_kwargs,  # type: ignore
+                torch.Size((sampling_batch_size,)),
+                **proposal_sampling_kwargs or {},  # type: ignore
             )  # type: ignore
 
         # SNPE-style rejection-sampling when the proposal is the neural net.

sbi/utils/sbiutils.py

@@ -679,7 +679,9 @@ def mcmc_transform(
                 # AttributeError -> Custom distribution that has no mean/std attribute.
                 warnings.warn(
                     """The passed prior has no mean or stddev attribute, estimating
-                    them from samples to build affimed standardizing transform."""
+                    them from samples to build affimed standardizing
+                    transform.""",
+                    stacklevel=2,
                 )
                 theta = prior.sample(torch.Size((num_prior_samples_for_zscoring,)))
                 prior_mean = theta.mean(dim=0).to(device)
@@ -719,8 +721,8 @@ def check_transform(
 
     assert torch.allclose(
         theta,
-        transform(theta_unconstrained),
-        atol=atol,  # type: ignore
+        transform(theta_unconstrained),  # type: ignore
+        atol=atol,
     ), "Original and re-transformed parameters must be close to each other."