Allow parsing of nested numbers from config (facebookresearch#400)

Summary:

Nested arrays were not directly parseable from the config. E.g., [12, [2, 4], 2, [2.0, 4]] would raise an exception.

If a nested array is detected, it will check whether the desired type is either a float or an int and if so, it'll evaluate the string as a number then recursively cast.

If the desired type is neither a float nor an int, it will raise a ValueException.

Differential Revision: D64138450

aepsych/config.py

@@ -182,11 +182,34 @@ def update(
                 self["common"][i] = self["experiment"][i]
             del self["experiment"]
 
-    def _str_to_list(self, v: str, element_type: _T = float) -> List[_T]:
+    def _str_to_list(
+        self, v: str, element_type: _T = float
+    ) -> List[_T] | List[List[_T]]:
         v = re.sub(r"\n ", ",", v)
         v = re.sub(r"(?<!,)\s+", ",", v)
         v = re.sub(r",]", "]", v)
         if re.search(r"^\[.*\]$", v, flags=re.DOTALL):
+            if len(re.findall(r"\[", v)) > 1:
+                if element_type in [float, int]:  # Easy to handle nested numbers
+
+                    def _nested_cast(v, element_type):
+                        result = []
+                        for item in v:
+                            if isinstance(item, list):
+                                # Recursively convert nested lists
+                                result.append(_nested_cast(item, element_type))
+                            else:
+                                # Convert individual values
+                                result.append(element_type(item))
+                        return result
+
+                    v = ast.literal_eval(v)
+                    return _nested_cast(v, element_type)
+
+                else:
+                    raise ValueError(
+                        "Nested list detected with an element_type that is hard to evaluate"
+                    )
             if v == "[]":  # empty list
                 return []
             else:

tests/test_config.py

@@ -186,7 +186,9 @@ def test_nonmonotonic_optimization_config_file(self):
         self.assertTrue(
             isinstance(strat.strat_list[1].generator, OptimizeAcqfGenerator)
         )
-        self.assertTrue(strat.strat_list[1].generator.acqf is qLogNoisyExpectedImprovement)
+        self.assertTrue(
+            strat.strat_list[1].generator.acqf is qLogNoisyExpectedImprovement
+        )
         self.assertTrue(
             set(strat.strat_list[1].generator.acqf_kwargs.keys()) == {"objective"}
         )
@@ -411,6 +413,31 @@ def test_warn_about_refit(self):
         with self.assertWarns(UserWarning):
             Strategy.from_config(config, "init_strat")
 
+    def test_nested_tensor(self):
+        points = [[0.25, 0.75], [0.5, 0.9]]
+        config_str = f"""
+                [common]
+                parnames = [par1, par2]
+
+                [par1]
+                par_type = continuous
+                lower_bound = 0
+                upper_bound = 1
+
+                [par2]
+                par_type = continuous
+                lower_bound = 0
+                upper_bound = 1
+
+                [SampleAroundPointsGenerator]
+                points = {points}
+        """
+        config = Config()
+        config.update(config_str=config_str)
+
+        config_points = config.gettensor("SampleAroundPointsGenerator", "points")
+        self.assertTrue(torch.all(config_points == torch.tensor(points)))
+
     def test_pairwise_probit_config(self):
         config_str = """
             [common]
@@ -581,7 +608,9 @@ def test_pairwise_opt_config(self):
         self.assertTrue(strat.strat_list[0].model is None)
 
         self.assertTrue(isinstance(strat.strat_list[1].model, PairwiseProbitModel))
-        self.assertTrue(strat.strat_list[1].generator.acqf is qLogNoisyExpectedImprovement)
+        self.assertTrue(
+            strat.strat_list[1].generator.acqf is qLogNoisyExpectedImprovement
+        )
         self.assertTrue(
             set(strat.strat_list[1].generator.acqf_kwargs.keys()) == {"objective"}
         )