Fix bug in _find_roots, add tol argument to _LM

CHANGELOG.rst

@@ -8,6 +8,13 @@ Changelog
 
 - The Wald test now supports robust covariance estimation.
 
+**Other changes:**
+
+- One can now pass the tolerance parameter `tol` to the optimization algorithm in
+  :func:`~ivmodels.tests.lagrange_multiplier.lagrange_multiplier_test` and
+  :func:`~ivmodels.tests.lagrange_multiplier.inverse_lagrange_multiplier_test` via the
+  `kwargs`.
+
 0.4.0 - 2024-08-08
 ------------------
 

ivmodels/summary.py

@@ -145,6 +145,7 @@ def fit(self, X, y, Z=None, C=None, *args, **kwargs):
         fit_intercept = self.kclass.fit_intercept
 
         for name in feature_names:
+
             idx = np.where(np.array(all_feature_names) == name)[0][0]
             mask = np.zeros(len(all_feature_names), dtype=bool)
             mask[idx] = True

ivmodels/tests/lagrange_multiplier.py

@@ -67,6 +67,8 @@ class _LM:
     gamma_0: list of str or np.ndarray of dimension (mw), optional, default=None
         Initial value for the minimization. If ``str``, must be one of "liml" or "zero".
         If ``None``, ``"liml"`` is used.
+    tol: float, optional, default=None
+        Tolerance for the optimization algorithm.
     """
 
     def __init__(
@@ -81,8 +83,8 @@ def __init__(
         W_proj=None,
         optimizer="bfgs",
         gamma_0=None,
+        tol=None,
     ):
-
         self.X = X
         self.y = y.reshape(-1, 1)
         self.W = W
@@ -116,6 +118,7 @@ def __init__(
         self.gamma_0 = ["liml"] if gamma_0 is None else gamma_0
         if isinstance(self.gamma_0, str):
             self.gamma_0 = [self.gamma_0]
+        self.tol = tol
 
     def liml(self, beta=None):
         """
@@ -299,7 +302,12 @@ def _derivative(gamma):
 
             results.append(
                 scipy.optimize.minimize(
-                    objective, jac=jac, hess=hess, x0=gamma_0, method=self.optimizer
+                    objective,
+                    jac=jac,
+                    hess=hess,
+                    x0=gamma_0,
+                    method=self.optimizer,
+                    tol=self.tol,
                 )
             )
 

ivmodels/utils.py

@@ -240,7 +240,7 @@ def _find_roots(f, a, b, tol, max_value, max_eval, n_points=50, max_depth=5):
     sgn = np.sign(b - a)
     if np.isinf(b):
         grid = np.ones(n_points) * a
-        grid[1:] += sgn * np.logspace(tol, np.log10(max_value), n_points - 1)
+        grid[1:] += sgn * np.logspace(np.log10(tol), np.log10(max_value), n_points - 1)
     else:
         grid = np.linspace(a, b, n_points)
 

tests/tests/test_lagrange_multiplier.py

@@ -36,7 +36,7 @@ def test__LM__init__(n, mx, mw, k):
         [
             np.all(np.isclose(lm1.__dict__[k], lm2.__dict__[k]))
             for k in lm1.__dict__.keys()
-            if k not in ["optimizer", "gamma_0"]
+            if k not in ["optimizer", "gamma_0", "tol"]
         ]
     )