refactor: more robust parsing of operators

pysr/sr.py

@@ -44,6 +44,7 @@
     _load_cluster_manager,
     jl_array,
     jl_deserialize,
+    jl_is_function,
     jl_serialize,
 )
 from .julia_import import SymbolicRegression, jl
@@ -1695,11 +1696,25 @@ def _run(
             optimize=self.weight_optimize,
         )
 
+        jl_binary_operators = []
+        jl_unary_operators = []
+        for input_list, output_list, name in [
+            (binary_operators, jl_binary_operators, "binary"),
+            (unary_operators, jl_unary_operators, "unary"),
+        ]:
+            for op in input_list:
+                jl_op = jl.seval(op)
+                if not jl_is_function(jl_op):
+                    raise ValueError(
+                        f"When building `{name}_operators`, `'{op}'` did not return a Julia function"
+                    )
+                output_list.append(jl_op)
+
         # Call to Julia backend.
         # See https://github.com/MilesCranmer/SymbolicRegression.jl/blob/master/src/OptionsStruct.jl
         options = SymbolicRegression.Options(
-            binary_operators=jl.seval(str(binary_operators).replace("'", "")),
-            unary_operators=jl.seval(str(unary_operators).replace("'", "")),
+            binary_operators=jl_array(jl_binary_operators, dtype=jl.Function),
+            unary_operators=jl_array(jl_unary_operators, dtype=jl.Function),
             bin_constraints=jl_array(bin_constraints),
             una_constraints=jl_array(una_constraints),
             complexity_of_operators=complexity_of_operators,

pysr/test/test.py

@@ -431,6 +431,16 @@ def test_load_model_simple(self):
         )
         np.testing.assert_allclose(model.predict(self.X), model3.predict(self.X))
 
+    def test_jl_function_error(self):
+        # TODO: Move this to better class
+        with self.assertRaises(ValueError) as cm:
+            PySRRegressor(unary_operators=["1"]).fit([[1]], [1])
+
+        self.assertIn(
+            "When building `unary_operators`, `'1'` did not return a Julia function",
+            str(cm.exception),
+        )
+
 
 def manually_create_model(equations, feature_names=None):
     if feature_names is None: