Fixed BestSoFar optimisers being broken.

optimistix/_root_find.py

@@ -92,10 +92,38 @@ def postprocess(self, fn, y, aux, args, options, state, tags, result):
 class _MinimToRoot(AbstractMinimiser, _ToRoot):
     solver: AbstractMinimiser
 
+    # Redeclare these three to work around the Equinox bug fixed here:
+    # https://github.com/patrick-kidger/equinox/pull/544
+    @property  # pyright: ignore
+    def rtol(self):
+        return self.solver.rtol
+
+    @property  # pyright: ignore
+    def atol(self):
+        return self.solver.atol
+
+    @property  # pyright: ignore
+    def norm(self):
+        return self.solver.norm
+
 
 class _LstsqToRoot(AbstractLeastSquaresSolver, _ToRoot):
     solver: AbstractLeastSquaresSolver
 
+    # Redeclare these three to work around the Equinox bug fixed here:
+    # https://github.com/patrick-kidger/equinox/pull/544
+    @property  # pyright: ignore
+    def rtol(self):
+        return self.solver.rtol
+
+    @property  # pyright: ignore
+    def atol(self):
+        return self.solver.atol
+
+    @property  # pyright: ignore
+    def norm(self):
+        return self.solver.norm
+
 
 @eqx.filter_jit
 def root_find(

optimistix/_solver/best_so_far.py

@@ -132,6 +132,20 @@ def __init__(self, solver: AbstractMinimiser[Y, tuple[Scalar, Aux], Any]):
     def _to_loss(self, y: Y, f: Scalar) -> Scalar:
         return f
 
+    # Redeclare these three to work around the Equinox bug fixed here:
+    # https://github.com/patrick-kidger/equinox/pull/544
+    @property  # pyright: ignore
+    def rtol(self):
+        return self.solver.rtol
+
+    @property  # pyright: ignore
+    def atol(self):
+        return self.solver.atol
+
+    @property  # pyright: ignore
+    def norm(self):
+        return self.solver.norm
+
 
 BestSoFarMinimiser.__init__.__doc__ = """**Arguments:**
 
@@ -158,6 +172,20 @@ def __init__(
     def _to_loss(self, y: Y, f: Out) -> Scalar:
         return sum_squares(f)
 
+    # Redeclare these three to work around the Equinox bug fixed here:
+    # https://github.com/patrick-kidger/equinox/pull/544
+    @property  # pyright: ignore
+    def rtol(self):
+        return self.solver.rtol
+
+    @property  # pyright: ignore
+    def atol(self):
+        return self.solver.atol
+
+    @property  # pyright: ignore
+    def norm(self):
+        return self.solver.norm
+
 
 BestSoFarLeastSquares.__init__.__doc__ = """**Arguments:**
 
@@ -181,6 +209,20 @@ def __init__(self, solver: AbstractRootFinder[Y, Out, tuple[Out, Aux], Any]):
     def _to_loss(self, y: Y, f: Out) -> Scalar:
         return sum_squares(f)
 
+    # Redeclare these three to work around the Equinox bug fixed here:
+    # https://github.com/patrick-kidger/equinox/pull/544
+    @property  # pyright: ignore
+    def rtol(self):
+        return self.solver.rtol
+
+    @property  # pyright: ignore
+    def atol(self):
+        return self.solver.atol
+
+    @property  # pyright: ignore
+    def norm(self):
+        return self.solver.norm
+
 
 BestSoFarRootFinder.__init__.__doc__ = """**Arguments:**
 
@@ -204,6 +246,20 @@ def __init__(self, solver: AbstractFixedPointSolver[Y, tuple[Y, Aux], Any]):
     def _to_loss(self, y: Y, f: Y) -> Scalar:
         return sum_squares((y**ω - f**ω).ω)
 
+    # Redeclare these three to work around the Equinox bug fixed here:
+    # https://github.com/patrick-kidger/equinox/pull/544
+    @property  # pyright: ignore
+    def rtol(self):
+        return self.solver.rtol
+
+    @property  # pyright: ignore
+    def atol(self):
+        return self.solver.atol
+
+    @property  # pyright: ignore
+    def norm(self):
+        return self.solver.norm
+
 
 BestSoFarFixedPoint.__init__.__doc__ = """**Arguments:**
 

tests/test_best_so_far.py

@@ -39,5 +39,6 @@ def fn(y, _):
         return 0.5 * (y - jnp.tanh(y + 1)) ** 2
 
     solver = optx.BFGS(rtol=1e-6, atol=1e-6)
+    solver = optx.BestSoFarMinimiser(solver)
     sol = optx.minimise(fn, solver, jnp.array(0.0))
     assert jnp.allclose(sol.value, 0.96118069, rtol=1e-5, atol=1e-5)