SearchLibrium 0.0.134__tar.gz → 0.0.136__tar.gz

{searchlibrium-0.0.134 → searchlibrium-0.0.136}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: SearchLibrium
-Version: 0.0.134
+Version: 0.0.136
 Summary: A Python package for econometric models driven by search
 Author: Mario Fernando Taco Morales, Alexander Paz, Prithvi Beeramole, Robert Burdett
 Author-email: Zeke Ahern <z.ahern@qut.edu.au>

{searchlibrium-0.0.134 → searchlibrium-0.0.136}/pyproject.toml

@@ -73,7 +73,7 @@ Documentation = "https://pypi.org/project/SearchLibrium/"
 realpython = "SearchLibrium.__main__:main"
 
 [tool.bumpver]
-current_version = "0.0.134"
+current_version = "0.0.136"
 version_pattern = "MAJOR.MINOR.PATCH"
 commit_message = "[skip ci] Bump version {old_version} -> {new_version}"
 commit = true

{searchlibrium-0.0.134 → searchlibrium-0.0.136}/src/SearchLibrium/MixedLogit.py

@@ -597,6 +597,14 @@ class MixedLogit(DiscreteChoiceModel):
             H = self.get_hessian(result['x'], self._get_loglik_gradient)
             result['hess_inv'] = np.linalg.inv(H)
         # }
+        else:
+            # ── Non-L-BFGS-B path (e.g. JAX): compute Hessian numerically
+            #     so post_process can extract standard errors
+            try:
+                H = self.get_hessian(result['x'], self._get_loglik_gradient, eps=1e-6)
+                result['hess_inv'] = np.linalg.inv(H)
+            except Exception:
+                pass
 
         if self.save_fitted_params:
             self.compute_fitted_params(self.y, self.p, self.panel_info, self.Br)
@@ -796,6 +804,21 @@ class MixedLogit(DiscreteChoiceModel):
 
             result = sp_min(
                 _obj, betas, jac=True, method=self.method, options=opts)
+
+            # ── JAX autograd Hessian for standard errors ───────────────
+            try:
+                hess_cache_key = _cache_key + ('hess',)
+                _compiled_hess = self._mxl_jit_cache.get(hess_cache_key)
+                if _compiled_hess is None:
+                    _compiled_hess = jax.jit(jax.hessian(_fn))
+                    self._mxl_jit_cache[hess_cache_key] = _compiled_hess
+                b_opt = jnp.array(result['x'], dtype=jnp.float64)
+                H = _compiled_hess(b_opt, X_jax, y_jax, pi_jax, draws_jax)
+                H_np = np.asarray(H, dtype=float)
+                result['hess_inv'] = np.linalg.inv(H_np)
+            except Exception:
+                pass
+
             return result
 
         except Exception as e:

{searchlibrium-0.0.134 → searchlibrium-0.0.136}/src/SearchLibrium/latent_class.py

@@ -42,7 +42,7 @@ class LatentClassMixedLogit:
         optimise_membership=True,
         membership_maxiter=50,
         l1_penalty=0.0,
-        l2_penalty=0.0,
+        l2_penalty=0.1,
     ):
         self.n_classes = int(n_classes)
         self.maxiter = int(maxiter)

{searchlibrium-0.0.134 → searchlibrium-0.0.136}/src/SearchLibrium/mixed_logit.py

@@ -619,6 +619,12 @@ class MixedLogit(DiscreteChoiceModel):
             H = get_hessian(result['x'], self._get_loglik_gradient)
             result['hess_inv'] = np.linalg.inv(H)
         # }
+        else:
+            try:
+                H = get_hessian(result['x'], self._get_loglik_gradient, eps=1e-6)
+                result['hess_inv'] = np.linalg.inv(H)
+            except Exception:
+                pass
 
         if self.save_fitted_params:
             self.compute_fitted_params(self.y, self.p, self.panel_info, self.Br)

{searchlibrium-0.0.134 → searchlibrium-0.0.136}/src/SearchLibrium/search.py

@@ -684,6 +684,10 @@ class Parameters:
         self.de_tol = de_tol
         self.de_polish = de_polish
 
+        # ── Regularisation (primarily for latent class) ──────────────
+        self.l1_penalty = kwargs.get('l1_penalty', 0.0)
+        self.l2_penalty = kwargs.get('l2_penalty', 0.1)
+
         self.intercept_opts = intercept_opts
         self.base_alt = base_alt
         self.val_share = val_share
@@ -4013,6 +4017,8 @@ class Search():
             random_state=seed if seed is not None else 0,
             optimise_membership=optimise_membership,
             membership_maxiter=100,
+            l1_penalty=getattr(self.param, 'l1_penalty', 0.0),
+            l2_penalty=getattr(self.param, 'l2_penalty', 0.0),
         )
 
         membership_vars = None
@@ -4216,6 +4222,65 @@ class Search():
     # }
 
 
+    ''' ---------------------------------------------------------- '''
+    ''' Function. Fit and Evaluate Latent Class Model              '''
+    ''' ---------------------------------------------------------- '''
+
+    def evaluate_lc(self, sol):
+    # {
+        sol = self.apply_constraints(sol)
+        as_vars = sol.get('asvars', [])
+        is_vars = sol.get('isvars', [])
+        asc_ind = sol.get('asc_ind', False)
+        bc_vars = self.define_bc_vars(sol)
+        all_vars = [var for var in self.param.varnames if var in (as_vars + is_vars)]
+
+        class_params_spec = sol.get('class_params_spec', None)
+        member_params_spec = sol.get('member_params_spec', None)
+        num_classes = getattr(self.param, 'num_classes', 2)
+
+        X = self.param.df[all_vars].values
+        y = self.param.choices
+        ids = self.param.choice_id if self.param.choice_id is not None else self.param.ind_id
+
+        alts = self.param.alt_var
+        if alts is None:
+            alts = np.ones(len(y), dtype=int)
+
+        model = self.fit_lcm(
+            X=X, y=y, varnames=all_vars,
+            class_params_spec=class_params_spec,
+            member_params_spec=member_params_spec,
+            num_classes=num_classes,
+            ids=ids,
+            transvars=bc_vars,
+            maxiter=self.param.maxiter,
+            gtol=self.param.gtol,
+            avail=self.param.avail,
+            weights=self.param.weights,
+            alts=alts,
+            base_alt=self.param.base_alt,
+        )
+
+        sol['model'] = model
+        sol['coeff'] = model.coeff_est
+        sol['model_n'] = 'latent_class'
+        converged = model.converged
+
+        aic = getattr(model, 'aic', float('inf'))
+        bic = getattr(model, 'bic', float('inf'))
+        loglik = getattr(model, 'loglik', float('-inf'))
+
+        mae = float('inf')
+
+        if getattr(self.param, 'verbose', False):
+            model.summarise()
+
+        tuple_ = (aic, bic, loglik, mae, as_vars, is_vars, {}, bc_vars, [], converged, sol)
+        return tuple_
+    # }
+
+
     ''' ---------------------------------------------------------- '''
     ''' Function. Fit and Evaluate Nested Logit Model              '''
     ''' ---------------------------------------------------------- '''
@@ -4582,6 +4647,10 @@ class Search():
     ''' ---------------------------------------------------------- '''
     def evaluate_model(self, sol):
     # {
+        # ── Latent class override: if param says latent, do latent ──
+        if getattr(self.param, 'latent_class', False):
+            return self.evaluate_lc(sol)
+
         model_n = sol.get('model_n', '')
 
         # ── Pre-fit collinearity / prerequisite check ────────────────

searchlibrium-0.0.136/src/SearchLibrium/version.txt

@@ -0,0 +1 @@
+0.0.136

{searchlibrium-0.0.134 → searchlibrium-0.0.136}/src/SearchLibrium.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: SearchLibrium
-Version: 0.0.134
+Version: 0.0.136
 Summary: A Python package for econometric models driven by search
 Author: Mario Fernando Taco Morales, Alexander Paz, Prithvi Beeramole, Robert Burdett
 Author-email: Zeke Ahern <z.ahern@qut.edu.au>

searchlibrium-0.0.134/src/SearchLibrium/version.txt

@@ -1 +0,0 @@
-0.0.134