SearchLibrium 0.0.83__tar.gz → 0.0.85__tar.gz

{searchlibrium-0.0.83 → searchlibrium-0.0.85}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: SearchLibrium
-Version: 0.0.83
+Version: 0.0.85
 Summary: A Python package for econometric models driven by search
 Author: Alexander Paz Prithvi Beeramole, Robert Burdett
 Author-email: Zeke Ahern <z.ahern@qut.edu.au>

{searchlibrium-0.0.83 → searchlibrium-0.0.85}/pyproject.toml

@@ -59,7 +59,7 @@ Homepage = "https://github.com/zahern/HypothesisX"
 realpython = "SearchLibrium.__main__:main"
 
 [tool.bumpver]
-current_version = "0.0.83"
+current_version = "0.0.85"
 version_pattern = "MAJOR.MINOR.PATCH"
 commit_message = "[skip ci] Bump version {old_version} -> {new_version}"
 commit = true

{searchlibrium-0.0.83 → searchlibrium-0.0.85}/src/SearchLibrium/__init__.py

@@ -55,7 +55,13 @@ def new_features():
            """)
 
 def get_version_from_pkg_info():
-    """Reads the version from the PKG-INFO file."""
+    """Reads the installed package version via importlib.metadata."""
+    try:
+        from importlib.metadata import version as _pkg_version
+        return _pkg_version("SearchLibrium")
+    except Exception:
+        pass
+    # Fallback: read from egg-info PKG-INFO (editable installs)
     pkg_info_path = os.path.join(os.path.dirname(__file__), "../SearchLibrium.egg-info/PKG-INFO")
     try:
         with open(pkg_info_path, "r") as f:
@@ -63,7 +69,8 @@ def get_version_from_pkg_info():
                 if line.startswith("Version:"):
                     return line.split(":")[1].strip()
     except FileNotFoundError:
-        return "0.0.32"
+        pass
+    return "unknown"
 
 __version__ = get_version_from_pkg_info()
 
@@ -86,7 +93,10 @@ try:
     from .rrm import RandomRegret
     from .mixedrrm import MixedRandomRegret
     from .ordered_logit import OrderedLogit, OrderedLogitLong
+    from .selection_models import BinaryProbit, HeckmanTwoStep
     from .latent_class import LatentClassMixedLogit
+    from .mdcev import MDCEVFitResult, MDCEVModel
+    from .multinomial_probit import MultinomialProbit
     from .RandomP import RandomParameters
     from .constraints_builder import ConstraintBuilder, create_constraints
     from .search import Parameters
@@ -102,21 +112,28 @@ except ImportError as e:
     from rrm import RandomRegret
     from mixedrrm import MixedRandomRegret
     from ordered_logit import OrderedLogit, OrderedLogitLong
+    from selection_models import BinaryProbit, HeckmanTwoStep
     from latent_class import LatentClassMixedLogit
+    from mdcev import MDCEVFitResult, MDCEVModel
+    from multinomial_probit import MultinomialProbit
     from RandomP import RandomParameters
     from constraints_builder import ConstraintBuilder, create_constraints
     from search import Parameters
     from call_meta import call_siman, call_harmony, call_search, estimate_ctrl
 try:
     from .main import print_ascii_art_logo
-except:
-    from main import print_ascii_art_logo
-    
+except Exception:
+    try:
+        from main import print_ascii_art_logo
+    except Exception:
+        print_ascii_art_logo = None
 
-try:
-    print_ascii_art_logo()
-except ImportError:
-    print("Error importing print_ascii_art_logo from main module. Continuing without logo.") 
+
+if print_ascii_art_logo is not None:
+    try:
+        print_ascii_art_logo()
+    except Exception:
+        print("SearchLibrium logo skipped; optional display dependencies are missing.")
 
 #print('loaded all')
 print('Welcome to SearchLibrium')

searchlibrium-0.0.85/src/SearchLibrium/mdcev.py

@@ -0,0 +1,344 @@
+"""MDCEV budget-allocation prototype for SearchLibrium.
+
+This module implements a compact translated-utility MDCEV-style allocator for
+continuous budget splits such as daily time-use or discretionary activity
+budgets. The implementation is forecasting-oriented: it provides a stable
+fitting heuristic from observed allocations together with an analytical
+budget-allocation solver based on the translated utility first-order
+conditions.
+
+The class is intended as a practical bridge between the current scalar budget
+models and a fuller MDCEV pipeline. It includes both a stable heuristic fit
+and a likelihood-based quasi-MLE refinement.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Iterable, Optional
+
+import numpy as np
+import pandas as pd
+from scipy.optimize import minimize
+
+
+def _as_2d_float(array_like) -> np.ndarray:
+    arr = np.asarray(array_like, dtype=float)
+    if arr.ndim == 1:
+        arr = arr.reshape(1, -1)
+    if arr.ndim != 2:
+        raise ValueError("Expected a 2D array of allocations")
+    return np.nan_to_num(arr, nan=0.0, posinf=0.0, neginf=0.0)
+
+
+@dataclass
+class MDCEVFitResult:
+    labels: list[str]
+    baseline_utility: np.ndarray
+    alpha: np.ndarray
+    gamma: np.ndarray
+    participation_rate: np.ndarray
+    mean_allocation: np.ndarray
+    mean_budget: float
+
+
+class MDCEVModel:
+    """Translated-utility MDCEV-style allocator.
+
+    Parameters are learned from observed budget shares using stable moment-based
+    heuristics, then predictions are produced by solving the translated-utility
+    KKT system with a bisection search on the shadow price.
+    """
+
+    def __init__(
+        self,
+        outside_good: Optional[int] = 0,
+        alpha_floor: float = 0.05,
+        alpha_cap: float = 0.95,
+        gamma_floor: float = 1e-3,
+        tol: float = 1e-9,
+    ):
+        self.outside_good = outside_good
+        self.alpha_floor = alpha_floor
+        self.alpha_cap = alpha_cap
+        self.gamma_floor = gamma_floor
+        self.tol = tol
+
+        self.labels_: list[str] | None = None
+        self.baseline_utility_: np.ndarray | None = None
+        self.alpha_: np.ndarray | None = None
+        self.gamma_: np.ndarray | None = None
+        self.fit_result_: MDCEVFitResult | None = None
+
+    def fit(self, allocations, labels: Optional[Iterable[str]] = None):
+        """Estimate baseline utility and satiation terms from observed allocations.
+
+        Parameters
+        ----------
+        allocations:
+            Matrix of observed budgets split across alternatives. Rows are
+            observations and columns are alternatives.
+        labels:
+            Optional alternative labels.
+        """
+        y = _as_2d_float(allocations)
+        n_obs, n_alt = y.shape
+        budgets = y.sum(axis=1)
+        if np.any(budgets < self.tol):
+            raise ValueError("Each observation must have a positive total budget")
+
+        labels_list = list(labels) if labels is not None else [f"alt_{i}" for i in range(n_alt)]
+        if len(labels_list) != n_alt:
+            raise ValueError("labels length must match number of alternatives")
+
+        positive = y > self.tol
+        participation = positive.mean(axis=0)
+        mean_allocation = y.mean(axis=0)
+        share = y.sum(axis=0) / np.clip(y.sum(), self.tol, None)
+
+        if self.outside_good is not None and 0 <= self.outside_good < n_alt:
+            ref_share = max(float(share[self.outside_good]), self.tol)
+            baseline = np.log(np.clip(share, self.tol, None)) - np.log(ref_share)
+            baseline[self.outside_good] = 0.0
+        else:
+            baseline = np.log(np.clip(share, self.tol, None))
+            baseline = baseline - baseline.mean()
+
+        gamma = np.full(n_alt, self.gamma_floor, dtype=float)
+        alpha = np.full(n_alt, 0.5, dtype=float)
+
+        for idx in range(n_alt):
+            pos_vals = y[positive[:, idx], idx]
+            if pos_vals.size == 0:
+                gamma[idx] = max(np.median(budgets) * 0.05, self.gamma_floor)
+                alpha[idx] = self.alpha_floor
+                baseline[idx] = min(baseline[idx], -8.0)
+                continue
+
+            median_pos = float(np.median(pos_vals))
+            mean_pos = float(np.mean(pos_vals))
+            std_pos = float(np.std(pos_vals))
+            cv_pos = std_pos / max(mean_pos, self.tol)
+
+            gamma[idx] = max(median_pos * max(1.0 - participation[idx], 0.1), self.gamma_floor)
+            raw_alpha = 0.2 + 0.6 * participation[idx] / (1.0 + cv_pos)
+            alpha[idx] = float(np.clip(raw_alpha, self.alpha_floor, self.alpha_cap))
+
+        if self.outside_good is not None and 0 <= self.outside_good < n_alt:
+            gamma[self.outside_good] = self.gamma_floor
+            alpha[self.outside_good] = max(alpha[self.outside_good], 0.8)
+
+        self.labels_ = labels_list
+        self.baseline_utility_ = baseline
+        self.alpha_ = alpha
+        self.gamma_ = gamma
+        self.fit_result_ = MDCEVFitResult(
+            labels=labels_list,
+            baseline_utility=baseline.copy(),
+            alpha=alpha.copy(),
+            gamma=gamma.copy(),
+            participation_rate=participation.copy(),
+            mean_allocation=mean_allocation.copy(),
+            mean_budget=float(np.mean(budgets)),
+        )
+        return self
+
+    def fit_mle(
+        self,
+        allocations,
+        labels: Optional[Iterable[str]] = None,
+        maxiter: int = 400,
+        l2_penalty: float = 1e-4,
+    ):
+        """Likelihood-based parameter refinement.
+
+        The objective is a Gaussian log-likelihood on log allocations around
+        translated-utility MDCEV deterministic predictions. This is a practical
+        quasi-MLE refinement that preserves the MDCEV budget constraint while
+        improving fit over pure moments.
+        """
+        self.fit(allocations, labels=labels)
+
+        y = _as_2d_float(allocations)
+        budgets = y.sum(axis=1)
+        n_alt = y.shape[1]
+
+        free_base_idx = [i for i in range(n_alt) if i != self.outside_good]
+
+        def _pack(base, alpha, gamma, sigma):
+            b = np.asarray(base, dtype=float)
+            a = np.asarray(alpha, dtype=float)
+            g = np.asarray(gamma, dtype=float)
+
+            p = []
+            p.extend(b[free_base_idx].tolist())
+            p.extend(np.log(np.clip((a - self.alpha_floor) / np.clip(self.alpha_cap - a, self.tol, None), self.tol, None)).tolist())
+            p.extend(np.log(np.clip(g, self.gamma_floor, None)).tolist())
+            p.append(np.log(max(float(sigma), 1e-3)))
+            return np.asarray(p, dtype=float)
+
+        def _unpack(theta):
+            theta = np.asarray(theta, dtype=float)
+            o = 0
+
+            base = self.baseline_utility_.copy()
+            for idx in free_base_idx:
+                base[idx] = theta[o]
+                o += 1
+            if self.outside_good is not None and 0 <= self.outside_good < n_alt:
+                base[self.outside_good] = 0.0
+
+            alpha_raw = theta[o:o + n_alt]
+            o += n_alt
+            alpha_sig = 1.0 / (1.0 + np.exp(-alpha_raw))
+            alpha = self.alpha_floor + (self.alpha_cap - self.alpha_floor) * alpha_sig
+
+            gamma_raw = theta[o:o + n_alt]
+            o += n_alt
+            gamma = np.maximum(np.exp(gamma_raw), self.gamma_floor)
+
+            sigma = max(np.exp(theta[o]), 1e-3)
+            return base, alpha, gamma, sigma
+
+        def _neg_loglike(theta):
+            base, alpha, gamma, sigma = _unpack(theta)
+
+            old_b, old_a, old_g = self.baseline_utility_, self.alpha_, self.gamma_
+            self.baseline_utility_, self.alpha_, self.gamma_ = base, alpha, gamma
+            try:
+                mu = np.zeros_like(y)
+                for i, b in enumerate(budgets):
+                    mu[i] = self._solve_budget(float(b), base)
+            finally:
+                self.baseline_utility_, self.alpha_, self.gamma_ = old_b, old_a, old_g
+
+            log_y = np.log(np.clip(y, self.tol, None))
+            log_mu = np.log(np.clip(mu, self.tol, None))
+            resid = log_y - log_mu
+            ll = -0.5 * resid.size * np.log(2.0 * np.pi * sigma * sigma)
+            ll -= 0.5 * np.sum((resid / sigma) ** 2)
+            ll -= l2_penalty * np.sum(theta * theta)
+            return -float(ll)
+
+        theta0 = _pack(self.baseline_utility_, self.alpha_, self.gamma_, sigma=0.5)
+        res = minimize(
+            _neg_loglike,
+            theta0,
+            method="L-BFGS-B",
+            options={"maxiter": int(maxiter), "ftol": 1e-9},
+        )
+
+        base, alpha, gamma, sigma = _unpack(res.x)
+        self.baseline_utility_ = base
+        self.alpha_ = alpha
+        self.gamma_ = gamma
+        self.noise_sigma_ = float(sigma)
+        self.mle_success_ = bool(res.success)
+        self.mle_message_ = str(res.message)
+        return self
+
+    def summary(self) -> pd.DataFrame:
+        if self.fit_result_ is None:
+            raise RuntimeError("Model must be fit before calling summary()")
+        result = self.fit_result_
+        return pd.DataFrame(
+            {
+                "alternative": result.labels,
+                "baseline_utility": result.baseline_utility,
+                "alpha": result.alpha,
+                "gamma": result.gamma,
+                "participation_rate": result.participation_rate,
+                "mean_allocation": result.mean_allocation,
+            }
+        )
+
+    def predict(self, budgets, utility_shift=None) -> np.ndarray:
+        """Predict deterministic budget allocations for one or more budgets.
+
+        Parameters
+        ----------
+        budgets:
+            Scalar or vector of total budgets.
+        utility_shift:
+            Optional additive utility adjustment. Can be shape ``(J,)`` or
+            ``(N, J)``.
+        """
+        self._check_fitted()
+        budgets_arr = np.asarray(budgets, dtype=float).reshape(-1)
+        shifts = self._prepare_utility_shift(utility_shift, len(budgets_arr))
+
+        predictions = np.zeros((len(budgets_arr), len(self.baseline_utility_)), dtype=float)
+        for row_idx, budget in enumerate(budgets_arr):
+            predictions[row_idx] = self._solve_budget(budget, self.baseline_utility_ + shifts[row_idx])
+        return predictions
+
+    def simulate(self, budgets, utility_shift=None, n_draws: int = 100, random_state: Optional[int] = None) -> np.ndarray:
+        """Simulate stochastic budget allocations with Gumbel utility shocks."""
+        self._check_fitted()
+        budgets_arr = np.asarray(budgets, dtype=float).reshape(-1)
+        shifts = self._prepare_utility_shift(utility_shift, len(budgets_arr))
+        rng = np.random.default_rng(random_state)
+
+        sims = np.zeros((n_draws, len(budgets_arr), len(self.baseline_utility_)), dtype=float)
+        for draw_idx in range(n_draws):
+            shocks = rng.gumbel(loc=0.0, scale=1.0, size=shifts.shape)
+            for row_idx, budget in enumerate(budgets_arr):
+                sims[draw_idx, row_idx] = self._solve_budget(
+                    budget,
+                    self.baseline_utility_ + shifts[row_idx] + shocks[row_idx],
+                )
+        return sims
+
+    def _prepare_utility_shift(self, utility_shift, n_rows: int) -> np.ndarray:
+        n_alt = len(self.baseline_utility_)
+        if utility_shift is None:
+            return np.zeros((n_rows, n_alt), dtype=float)
+
+        shift_arr = np.asarray(utility_shift, dtype=float)
+        if shift_arr.ndim == 1:
+            if shift_arr.shape[0] != n_alt:
+                raise ValueError("utility_shift has the wrong number of alternatives")
+            return np.repeat(shift_arr.reshape(1, -1), n_rows, axis=0)
+        if shift_arr.shape != (n_rows, n_alt):
+            raise ValueError("utility_shift must have shape (J,) or (N, J)")
+        return shift_arr
+
+    def _solve_budget(self, budget: float, utility_index: np.ndarray) -> np.ndarray:
+        if budget <= self.tol:
+            return np.zeros(len(self.baseline_utility_), dtype=float)
+
+        weights = np.exp(np.clip(utility_index, -40.0, 40.0))
+
+        def alloc_for_lambda(lam: float) -> np.ndarray:
+            lam = max(lam, self.tol)
+            power = 1.0 / np.clip(1.0 - self.alpha_, self.tol, None)
+            raw = np.power(weights / lam, power) - self.gamma_
+            return np.maximum(raw, 0.0)
+
+        lo = self.tol
+        hi = max(np.max(weights), 1.0)
+        while alloc_for_lambda(hi).sum() > budget:
+            hi *= 2.0
+
+        for _ in range(80):
+            mid = 0.5 * (lo + hi)
+            if alloc_for_lambda(mid).sum() > budget:
+                lo = mid
+            else:
+                hi = mid
+
+        allocation = alloc_for_lambda(hi)
+        total = allocation.sum()
+        if total > self.tol:
+            allocation *= budget / total
+        elif self.outside_good is not None and 0 <= self.outside_good < len(allocation):
+            allocation[self.outside_good] = budget
+
+        residual = budget - allocation.sum()
+        if self.outside_good is not None and 0 <= self.outside_good < len(allocation) and residual > self.tol:
+            allocation[self.outside_good] += residual
+        return allocation
+
+    def _check_fitted(self):
+        if self.fit_result_ is None or self.baseline_utility_ is None:
+            raise RuntimeError("Model must be fit before prediction")

searchlibrium-0.0.85/src/SearchLibrium/selection_models.py

@@ -0,0 +1,268 @@
+import math
+from dataclasses import dataclass
+
+import numpy as np
+import pandas as pd
+from scipy.optimize import minimize
+from scipy.stats import norm, t as student_t
+
+try:
+    import jax
+    import jax.numpy as jnp
+    from jax.scipy.special import ndtr as jax_ndtr
+except ImportError:  # pragma: no cover
+    jax = None
+    jnp = None
+    jax_ndtr = None
+
+try:
+    from ._choice_model import DiscreteChoiceModel
+except ImportError:
+    from _choice_model import DiscreteChoiceModel
+
+
+class BinaryProbit(DiscreteChoiceModel):
+    """Binary probit estimated with JAX autodiff and scipy L-BFGS-B."""
+
+    def __init__(self, _jax=False):
+        super(BinaryProbit, self).__init__(_jax)
+        self.descr = "Binary Probit"
+        self.result = None
+        self._X_design = None
+
+    def setup(self, X, y, varnames=None, fit_intercept=True):
+        X = np.asarray(X)
+        y = np.asarray(y).reshape(-1)
+        if varnames is None:
+            varnames = [f"x{i}" for i in range(X.shape[1])]
+        self.X = X
+        self.y = y
+        self.varnames = np.asarray(varnames, dtype="<U64")
+        self.fit_intercept = bool(fit_intercept)
+        self.sample_size = int(X.shape[0])
+        if self.fit_intercept:
+            self._X_design = np.column_stack([np.ones((X.shape[0], 1)), X])
+            self._design_names = np.asarray(["intercept", *self.varnames], dtype="<U64")
+        else:
+            self._X_design = X.copy()
+            self._design_names = self.varnames.copy()
+        return self
+
+    def _negloglik_jax(self, params, X, y):
+        xb = X @ params
+        p = jnp.clip(jax_ndtr(xb), 1e-10, 1.0 - 1e-10)
+        ll = y * jnp.log(p) + (1.0 - y) * jnp.log(1.0 - p)
+        return -jnp.sum(ll)
+
+    def fit(self, disp=False, **fit_kwargs):
+        if jax is None or jnp is None or jax_ndtr is None:
+            raise ImportError("JAX is required for BinaryProbit")
+
+        X = jnp.asarray(self._X_design)
+        y = jnp.asarray(self.y)
+        init = np.zeros(X.shape[1], dtype=float)
+
+        val_grad = jax.jit(jax.value_and_grad(self._negloglik_jax))
+
+        def _obj(params_np):
+            val, grad = val_grad(jnp.asarray(params_np), X, y)
+            return float(val), np.asarray(grad, dtype=float)
+
+        res = minimize(
+            fun=lambda p: _obj(p)[0],
+            x0=init,
+            jac=lambda p: _obj(p)[1],
+            method="L-BFGS-B",
+            options={"disp": bool(disp), "maxiter": int(fit_kwargs.pop("maxiter", 1000))},
+        )
+        self.result = res
+        self.coeff_names = self._design_names.copy()
+        self.coeff_est = np.asarray(res.x, dtype=float)
+        self.loglik = float(-res.fun)
+        self.converged = bool(res.success)
+        self.total_fun_eval = int(getattr(res, "nfev", 0))
+
+        hess_inv = getattr(res, "hess_inv", None)
+        if hess_inv is not None:
+            if hasattr(hess_inv, "todense"):
+                cov = np.asarray(hess_inv.todense(), dtype=float)
+            else:
+                cov = np.asarray(hess_inv, dtype=float)
+            stderr = np.sqrt(np.clip(np.diag(cov), 1e-12, None))
+        else:
+            stderr = np.full_like(self.coeff_est, np.nan, dtype=float)
+
+        self.stderr = stderr
+        self.zvalues = self.coeff_est / np.where(stderr > 0, stderr, np.nan)
+        self.pvalues = 2.0 * (1.0 - norm.cdf(np.abs(self.zvalues)))
+        k = len(self.coeff_est)
+        n = max(int(self.sample_size), 1)
+        self.aic = float(2 * k - 2 * self.loglik)
+        self.bic = float(k * np.log(n) - 2 * self.loglik)
+        return res
+
+    def predict_proba(self, X=None):
+        if self.coeff_est is None:
+            raise RuntimeError("BinaryProbit must be fit before prediction")
+        X_arr = self.X if X is None else np.asarray(X)
+        if self.fit_intercept:
+            X_arr = np.column_stack([np.ones((X_arr.shape[0], 1)), X_arr])
+        xb = X_arr @ self.coeff_est
+        return norm.cdf(xb)
+
+    def summary_frame(self):
+        if self.coeff_est is None:
+            return pd.DataFrame()
+        return pd.DataFrame({
+            "coef": self.coeff_est,
+            "stderr": self.stderr,
+            "z": self.zvalues,
+            "pvalue": self.pvalues,
+        }, index=self.coeff_names)
+
+
+@dataclass
+class _OLSResult:
+    params: pd.Series
+    bse: pd.Series
+    tvalues: pd.Series
+    pvalues: pd.Series
+    llf: float
+
+
+class HeckmanTwoStep(DiscreteChoiceModel):
+    """Heckman selection model using JAX probit + closed-form OLS second stage."""
+
+    def __init__(self, _jax=False):
+        super(HeckmanTwoStep, self).__init__(_jax)
+        self.descr = "Heckman Two-Step"
+        self.selection_result = None
+        self.outcome_result = None
+        self.params_table = pd.DataFrame()
+
+    def setup(
+        self,
+        selection_X,
+        selection_y,
+        outcome_X,
+        outcome_y,
+        selection_varnames=None,
+        outcome_varnames=None,
+        fit_intercept=True,
+    ):
+        selection_X = np.asarray(selection_X)
+        selection_y = np.asarray(selection_y).reshape(-1)
+        outcome_X = np.asarray(outcome_X)
+        outcome_y = np.asarray(outcome_y).reshape(-1)
+        if selection_varnames is None:
+            selection_varnames = [f"s{i}" for i in range(selection_X.shape[1])]
+        if outcome_varnames is None:
+            outcome_varnames = [f"o{i}" for i in range(outcome_X.shape[1])]
+        self.selection_X = selection_X
+        self.selection_y = selection_y
+        self.outcome_X = outcome_X
+        self.outcome_y = outcome_y
+        self.selection_varnames = np.asarray(selection_varnames, dtype="<U64")
+        self.outcome_varnames = np.asarray(outcome_varnames, dtype="<U64")
+        self.fit_intercept = bool(fit_intercept)
+        self.sample_size = int(selection_X.shape[0])
+        return self
+
+    def fit(self, disp=False, **fit_kwargs):
+        sel_X = np.asarray(self.selection_X, dtype=float)
+        out_X = np.asarray(self.outcome_X, dtype=float)
+        if self.fit_intercept:
+            sel_X = np.column_stack([np.ones((sel_X.shape[0], 1)), sel_X])
+            out_X = np.column_stack([np.ones((out_X.shape[0], 1)), out_X])
+
+        probit_model = BinaryProbit(_jax=True)
+        sel_names = (["intercept"] if self.fit_intercept else []) + list(self.selection_varnames)
+        probit_model.setup(sel_X[:, 1:] if self.fit_intercept else sel_X,
+                           self.selection_y,
+                           varnames=sel_names[1:] if self.fit_intercept else sel_names,
+                           fit_intercept=self.fit_intercept)
+        probit_model.fit(disp=disp, **fit_kwargs)
+
+        xb = sel_X @ probit_model.coeff_est
+        mills = norm.pdf(xb) / np.clip(norm.cdf(xb), 1e-10, None)
+
+        mask = self.selection_y == 1
+        out_design = np.column_stack([out_X[mask], mills[mask]])
+        out_y = self.outcome_y[mask]
+
+        xtx = out_design.T @ out_design
+        xtx_inv = np.linalg.pinv(xtx)
+        beta = xtx_inv @ (out_design.T @ out_y)
+        resid = out_y - out_design @ beta
+        dof = max(out_design.shape[0] - out_design.shape[1], 1)
+        sigma2 = float((resid @ resid) / dof)
+        cov = sigma2 * xtx_inv
+        se = np.sqrt(np.clip(np.diag(cov), 1e-12, None))
+        tvals = beta / np.where(se > 0, se, np.nan)
+        pvals = 2.0 * (1.0 - student_t.cdf(np.abs(tvals), df=dof))
+        ll_ols = -0.5 * out_design.shape[0] * (math.log(2.0 * math.pi * sigma2) + 1.0)
+
+        out_names = (["intercept"] if self.fit_intercept else []) + list(self.outcome_varnames) + ["IMR"]
+        ols = _OLSResult(
+            params=pd.Series(beta, index=out_names),
+            bse=pd.Series(se, index=out_names),
+            tvalues=pd.Series(tvals, index=out_names),
+            pvalues=pd.Series(pvals, index=out_names),
+            llf=float(ll_ols),
+        )
+
+        self.selection_result = probit_model
+        self.outcome_result = ols
+        self.loglik = float(probit_model.loglik + ll_ols)
+        total_k = len(probit_model.coeff_est) + len(beta)
+        self.aic = float(2 * total_k - 2 * self.loglik)
+        self.bic = float(total_k * np.log(max(self.sample_size, 1)) - 2 * self.loglik)
+        self.converged = bool(probit_model.converged)
+
+        selection_tbl = pd.DataFrame({
+            "coef": probit_model.coeff_est,
+            "stderr": probit_model.stderr,
+            "z": probit_model.zvalues,
+            "pvalue": probit_model.pvalues,
+        }, index=probit_model.coeff_names)
+        outcome_tbl = pd.DataFrame({
+            "coef": ols.params,
+            "stderr": ols.bse,
+            "z": ols.tvalues,
+            "pvalue": ols.pvalues,
+        })
+        self.params_table = pd.concat(
+            {"selection": selection_tbl, "outcome": outcome_tbl},
+            names=["equation", "term"],
+        )
+
+        coeff_names = [f"selection::{name}" for name in selection_tbl.index]
+        coeff_names += [f"outcome::{name}" for name in outcome_tbl.index]
+        self.coeff_names = np.asarray(coeff_names, dtype="<U128")
+        self.coeff_est = np.concatenate([selection_tbl["coef"].values, outcome_tbl["coef"].values])
+        self.stderr = np.concatenate([selection_tbl["stderr"].values, outcome_tbl["stderr"].values])
+        self.zvalues = np.concatenate([selection_tbl["z"].values, outcome_tbl["z"].values])
+        self.pvalues = np.concatenate([selection_tbl["pvalue"].values, outcome_tbl["pvalue"].values])
+        return {"probit": probit_model, "ols": ols}
+
+    def predict_selection_proba(self, X=None):
+        if self.selection_result is None:
+            raise RuntimeError("HeckmanTwoStep must be fit before prediction")
+        X_arr = self.selection_X if X is None else np.asarray(X)
+        return self.selection_result.predict_proba(X_arr)
+
+    def predict_outcome(self, X=None, selection_probability=None):
+        if self.outcome_result is None:
+            raise RuntimeError("HeckmanTwoStep must be fit before prediction")
+        X_arr = self.outcome_X if X is None else np.asarray(X)
+        if self.fit_intercept:
+            X_arr = np.column_stack([np.ones((X_arr.shape[0], 1)), X_arr])
+        if selection_probability is None:
+            selection_probability = np.clip(self.predict_selection_proba(), 1e-10, 1 - 1e-10)
+        xb = norm.ppf(np.clip(selection_probability, 1e-10, 1 - 1e-10))
+        imr = norm.pdf(xb) / np.clip(norm.cdf(xb), 1e-10, None)
+        X_aug = np.column_stack([X_arr, imr])
+        return X_aug @ self.outcome_result.params.values
+
+    def summary_frame(self):
+        return self.params_table.copy()

{searchlibrium-0.0.83 → searchlibrium-0.0.85}/src/SearchLibrium/siman.py

@@ -1102,7 +1102,7 @@ class SA(Search):
         # {
             if overall_best_solution is None or \
                 is_better(self.best_sol.obj(0), overall_best_solution.obj(0), self.param.sign_crit(0)):
-                overall_best_solution = self.best_sol  # Update overall best solution
+                overall_best_solution = self.copy_solution(self.best_sol)  # Update overall best solution (deep copy to prevent overwriting)
             elif overall_best_solution is not None and \
                 is_worse(self.best_sol.obj(0), overall_best_solution.obj(0), self.param.sign_crit(0)):
                 self.update_best(overall_best_solution)  # Revise best solution of current SA solver

{searchlibrium-0.0.83 → searchlibrium-0.0.85}/src/SearchLibrium/threshold.py

@@ -448,7 +448,7 @@ class TA(Search):
         # {
             if overall_best_solution is None or \
                 is_better(self.best_sol.obj(0), overall_best_solution.obj(0), self.param.sign_crit(0)):
-                overall_best_solution = self.best_sol  # Update overall best solution
+                overall_best_solution = self.copy_solution(self.best_sol)  # Update overall best solution (deep copy to prevent overwriting)
             elif overall_best_solution is not None and \
                 is_worse(self.best_sol.obj(0), overall_best_solution.obj(0), self.param.sign_crit(0)):
                 self.update_best(overall_best_solution)  # Revise best solution of current TA solver

searchlibrium-0.0.85/src/SearchLibrium/version.txt

@@ -0,0 +1 @@
+0.0.85

{searchlibrium-0.0.83 → searchlibrium-0.0.85}/src/SearchLibrium.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: SearchLibrium
-Version: 0.0.83
+Version: 0.0.85
 Summary: A Python package for econometric models driven by search
 Author: Alexander Paz Prithvi Beeramole, Robert Burdett
 Author-email: Zeke Ahern <z.ahern@qut.edu.au>

{searchlibrium-0.0.83 → searchlibrium-0.0.85}/src/SearchLibrium.egg-info/SOURCES.txt

@@ -17,6 +17,7 @@ src/SearchLibrium/harmony.py
 src/SearchLibrium/latent_class.py
 src/SearchLibrium/main.py
 src/SearchLibrium/main_debug.py
+src/SearchLibrium/mdcev.py
 src/SearchLibrium/misc.py
 src/SearchLibrium/mixed_logit.py
 src/SearchLibrium/mixed_nested.py
@@ -28,6 +29,7 @@ src/SearchLibrium/ordered_logit.py
 src/SearchLibrium/ordered_logit_mixed.py
 src/SearchLibrium/rrm.py
 src/SearchLibrium/search.py
+src/SearchLibrium/selection_models.py
 src/SearchLibrium/setup.py
 src/SearchLibrium/siman.py
 src/SearchLibrium/threshold.py

searchlibrium-0.0.83/src/SearchLibrium/version.txt

@@ -1 +0,0 @@
-0.0.83