Minor refactor of pypfilt.resample

src/pypfilt/resample.py

@@ -39,9 +39,8 @@ def post_regularise(ctx, px, new_px):
     # List parameters in the order they appear in the state vector.
     smooth_fields = [n for n in field_names if n in smooth_fields]
     num_params = len(smooth_fields)
-    p_ixs = np.array(range(num_params))
 
-    if len(p_ixs) == 0:
+    if num_params == 0:
         logger.debug("Post-RPF: no parameters to resample")
         return
 
@@ -51,21 +50,20 @@ def post_regularise(ctx, px, new_px):
     # Check for parameters that are constant (or nearly so) for all particles.
     # These parameters must be ignored or the covariance matrix will not be
     # positive definite, and the Cholesky decomposition will fail.
-    p_range = np.ptp(x[:, p_ixs], axis=0)
+    p_range = np.ptp(x, axis=0)
     toln = ctx.params['resample']['reg_toln']
     good = p_range >= toln
     if not np.all(good):
         bad = np.logical_not(good)
         msg = "Post-RPF found {} constant parameter(s) at {}".format(
-            sum(bad), p_ixs[np.where(bad)])
+            sum(bad), np.array(smooth_fields)[bad])
         logger.debug(msg)
 
         # Update the variables related to these parameters.
         smooth_fields = [name for name in np.array(smooth_fields)[good]]
         num_params = len(smooth_fields)
-        p_ixs = np.array(range(num_params))
 
-        if len(p_ixs) == 0:
+        if num_params == 0:
             logger.debug("Post-RPF: no non-constant parameters to resample")
             return
 
@@ -75,8 +73,7 @@ def post_regularise(ctx, px, new_px):
     # Use a bandwidth that is half that of the optimal bandwidth for a
     # Gaussian kernel (when the underlying density is Gaussian with unit
     # covariance), to handle multi-model densities.
-    npar = len(p_ixs)
-    h = 0.5 * (4 / (count * (npar + 2))) ** (1 / (npar + 4))
+    h = 0.5 * (4 / (count * (num_params + 2))) ** (1 / (num_params + 4))
 
     # Calculate the Cholesky decomposition of the parameter covariance
     # matrix V, which is used to transform independent normal samples
@@ -107,7 +104,7 @@ def post_regularise(ctx, px, new_px):
             return
 
     # Sample the multivariate normal with covariance V and mean of zero.
-    std_samples = rnd.normal(size=(npar, count))
+    std_samples = rnd.normal(size=(num_params, count))
     scaled_samples = np.transpose(np.dot(a_mat, h * std_samples))
 
     # Add the sampled noise and clip to respect parameter bounds.