Store parameter bounds by name, not index

src/pypfilt/config.py

@@ -425,9 +425,6 @@ def make_params_fn(scen_data, settings: _Settings,
     overrides = {k: v for (k, v) in settings.params.items()
                  if k not in ignore}
 
-    descr = {field[0]: ix for (ix, field) in
-             enumerate(settings.model.state_fields())}
-
     def params_fn():
         params = default_params(settings.model,
                                 settings.time,
@@ -456,10 +453,9 @@ def make_params_fn(scen_data, settings: _Settings,
 
         if settings.bounds:
             for (name, info) in settings.bounds.items():
-                # NOTE: apply parameter bounds.
-                param_ix = descr[name]
-                params['model']['param_min'][param_ix] = info['min']
-                params['model']['param_max'][param_ix] = info['max']
+                # NOTE: apply parameter bounds by name.
+                params['model']['param_bounds'][name] = (info['min'],
+                                                         info['max'])
 
         # Allow priors to specify an RNG sampling function and arguments.
         if settings.priors:

src/pypfilt/params.py

@@ -16,10 +16,7 @@ def default_params(model, time_scale, max_days, px_count, prng_seed):
     :param px_count: The number of particles.
     :param prng_seed: The seed for the pseudo-random number generators.
     """
-    field_names = [field[0] for field in model.state_fields()]
     bounds = model.bounds()
-    p_min = [bounds[name][0] for name in field_names]
-    p_max = [bounds[name][1] for name in field_names]
     params = {
         'resample': {
             # Resample when the effective number of particles is 25%.
@@ -80,10 +77,8 @@ def default_params(model, time_scale, max_days, px_count, prng_seed):
             'until': None,
         },
         'model': {
-            # The lower bounds for each model parameter.
-            'param_min': np.array(p_min),
-            # The upper bounds for each model parameter.
-            'param_max': np.array(p_max),
+            # The lower and upper bounds for each model parameter.
+            'param_bounds': bounds,
             # The model prior distributions.
             'prior': {},
         },

src/pypfilt/resample.py

@@ -111,12 +111,8 @@ def post_regularise(ctx, px, new_px):
     scaled_samples = np.transpose(np.dot(a_mat, h * std_samples))
 
     # Add the sampled noise and clip to respect parameter bounds.
-    smooth_ixs = [ix for (ix, name) in enumerate(field_names)
-                  if name in smooth_fields]
     for (ix, name) in enumerate(smooth_fields):
-        param_ix = smooth_ixs[ix]
-        min_val = ctx.params['model']['param_min'][param_ix]
-        max_val = ctx.params['model']['param_max'][param_ix]
+        (min_val, max_val) = ctx.params['model']['param_bounds'][name]
         new_px['state_vec'][name] = np.clip(
             new_px['state_vec'][name] + scaled_samples[:, ix],
             min_val, max_val)