Tidy up bridging densities and pCN sampler

Author: alexgdebeer

deep_tensor/bridging_densities/bridge.py

@@ -6,21 +6,16 @@
 
 
 class Bridge(abc.ABC):
-
-    @property 
-    @abc.abstractmethod 
-    def is_adaptive(self) -> bool:
-        return
 
     @property
     @abc.abstractmethod
     def is_last(self) -> bool:
-        return
+        pass
 
     @property 
     @abc.abstractmethod
     def params_dict(self) -> Dict:
-        return
+        pass
 
     @property 
     def n_layers(self) -> int:
@@ -77,7 +72,7 @@ def _get_ratio_func(
             function evaluated for each sample.
             
         """
-        return
+        pass
 
     @abc.abstractmethod
     def _compute_log_weights(
@@ -112,7 +107,7 @@ def _compute_log_weights(
             previous bridging density evaluated at each sample.
         
         """
-        return
+        pass
 
     def _set_init(self, neglogliks: Tensor) -> None:
         """Computes the properties of the initial bridging density.

deep_tensor/bridging_densities/tempering.py

@@ -40,7 +40,7 @@ def max_layers(self, value: int) -> None:
     def is_last(self) -> bool:
         max_layers_reached = self.n_layers == self.max_layers
         final_beta_reached = (self.betas[self.n_layers-1] - 1.0).abs() < 1e-6
-        return max_layers_reached or final_beta_reached
+        return bool(max_layers_reached or final_beta_reached)
 
     @property
     def params_dict(self) -> Dict:

deep_tensor/debiasing/mcmc.py

@@ -134,8 +134,8 @@ def run_dirt_pcn(
     dirt: AbstractDIRT,
     n: int,
     dt: float = 2.0,
-    y_obs: Tensor | None = None,
     x0: Tensor | None = None,
+    ys: Tensor | None = None,
     subset: str = "first",
     verbose: bool = True
 ) -> MCMCResult:
@@ -157,8 +157,6 @@ def run_dirt_pcn(
         unnormalised) target density at a given sample.
     dirt:
         A previously-constructed DIRT object.
-    y_obs:
-        A tensor containing the observations.
     n: 
         The length of the Markov chain to construct.
     dt:
@@ -169,6 +167,12 @@ def run_dirt_pcn(
         be applied to it to generate the starting location for sampling 
         from the pullback of the target density. Otherwise, the mean of 
         the reference density will be used.
+    ys:
+        A tensor containing a set of values to condition on.
+    subset:
+        If `ys` are passed in, whether they correspond to the first 
+        $k$ variables (`subset='first'`) or the final $k$ variables 
+        (`subset='last'`).
     verbose:
         Whether to print diagnostic information during the sampling 
         process.
@@ -218,43 +222,43 @@ def run_dirt_pcn(
         msg = "Stepsize must be positive."
         raise Exception(msg)
 
-    if y_obs is not None:
-
-        y_obs = torch.atleast_2d(y_obs)
-        dim = dirt.dim - y_obs.shape[1]
+    if ys is None:
+        
+        dim = dirt.dim
 
         def negloglik_pullback(rs: Tensor) -> Tensor:
             """Returns the difference between the negative logarithm of the 
             pullback of the target function under the DIRT mapping and the 
             negative log-prior density.
             """
             rs = torch.atleast_2d(rs)
-            neglogfr = dirt.eval_cirt_pullback(potential, y_obs, rs, subset=subset)
+            neglogfr = dirt.eval_irt_pullback(potential, rs, subset=subset)
             neglogref = dirt.reference.eval_potential(rs)[0]
             return neglogfr - neglogref
-    
+        
         def irt_func(rs: Tensor) -> Tensor:
             rs = torch.atleast_2d(rs)
-            ms = dirt.eval_cirt(y_obs, rs, subset=subset)[0]
+            ms = dirt.eval_irt(rs, subset=subset)[0]
             return ms
 
     else:
 
-        dim = dirt.dim
+        ys = torch.atleast_2d(ys)
+        dim = dirt.dim - ys.shape[1]
 
         def negloglik_pullback(rs: Tensor) -> Tensor:
             """Returns the difference between the negative logarithm of the 
             pullback of the target function under the DIRT mapping and the 
             negative log-prior density.
             """
             rs = torch.atleast_2d(rs)
-            neglogfr = dirt.eval_irt_pullback(potential, rs, subset=subset)
+            neglogfr = dirt.eval_cirt_pullback(potential, ys, rs, subset=subset)
             neglogref = dirt.reference.eval_potential(rs)[0]
             return neglogfr - neglogref
-        
+    
         def irt_func(rs: Tensor) -> Tensor:
             rs = torch.atleast_2d(rs)
-            ms = dirt.eval_irt(rs, subset=subset)[0]
+            ms = dirt.eval_cirt(ys, rs, subset=subset)[0]
             return ms
 
     res = _run_irt_pcn(