Fix the coercion of scores to floats in the optimizer

mlos_bench/mlos_bench/optimizers/mlos_core_optimizer.py

@@ -104,20 +104,19 @@ def bulk_register(self,
         df_scores = self._adjust_signs_df(
             pd.DataFrame([{} if score is None else score for score in scores]))
 
-        opt_targets = list(self._opt_targets)
         if status is not None:
             # Select only the completed trials, set scores for failed trials to +inf.
             df_status = pd.Series(status)
             # TODO: Be more flexible with values used for failed trials (not just +inf).
             # Issue: https://github.com/microsoft/MLOS/issues/523
-            df_scores.loc[df_status != Status.SUCCEEDED, opt_targets] = float("inf")
+            df_scores[df_status != Status.SUCCEEDED] = float("inf")
             df_status_completed = df_status.apply(Status.is_completed)
             df_configs = df_configs[df_status_completed]
             df_scores = df_scores[df_status_completed]
 
         # TODO: Specify (in the config) which metrics to pass to the optimizer.
         # Issue: https://github.com/microsoft/MLOS/issues/745
-        self._opt.register(configs=df_configs, scores=df_scores[opt_targets].astype(float))
+        self._opt.register(configs=df_configs, scores=df_scores)
 
         if _LOG.isEnabledFor(logging.DEBUG):
             (score, _) = self.get_best_observation()
@@ -127,11 +126,12 @@ def bulk_register(self,
 
     def _adjust_signs_df(self, df_scores: pd.DataFrame) -> pd.DataFrame:
         """
-        In-place adjust the signs of the scores for MINIMIZATION problem.
+        Coerce optimization target scores to floats
+        and adjust the signs for MINIMIZATION problem.
         """
-        for (opt_target, opt_dir) in self._opt_targets.items():
-            df_scores[opt_target] *= opt_dir
-        return df_scores
+        # NOTE: We use `.astype()` instead of `.apply(pd.to_numeric)`
+        # to catch incorrect values early.
+        return df_scores[list(self._opt_targets)].astype(float) * self._opt_targets.values()
 
     def _to_df(self, configs: Sequence[Dict[str, TunableValue]]) -> pd.DataFrame:
         """

mlos_bench/mlos_bench/tests/optimizers/mlos_core_opt_df_test.py

@@ -28,6 +28,10 @@ def mlos_core_optimizer(tunable_groups: TunableGroups) -> MlosCoreOptimizer:
         'optimizer_type': 'FLAML',
         'max_suggestions': 10,
         'seed': SEED,
+        'optimization_targets': {
+            'latency': 'min',
+            'throughput': 'max',
+        },
     }
     return MlosCoreOptimizer(tunable_groups, test_opt_config)
 
@@ -81,3 +85,38 @@ def test_df(mlos_core_optimizer: MlosCoreOptimizer, mock_configs: List[dict]) ->
             'vmSize': 'Standard_B2s',
         },
     ]
+
+
+def test_df_str(mlos_core_optimizer: MlosCoreOptimizer, mock_configs: List[dict]) -> None:
+    """
+    Test `MlosCoreOptimizer._to_df()` type coercion on tunables with string values.
+    """
+    df_config_orig = mlos_core_optimizer._to_df(mock_configs)
+    df_config_str = mlos_core_optimizer._to_df([
+        {key: str(val) for (key, val) in config.items()}
+        for config in mock_configs
+    ])
+    assert df_config_orig.equals(df_config_str)
+
+
+def test_adjust_signs_df(mlos_core_optimizer: MlosCoreOptimizer) -> None:
+    """
+    Test `MlosCoreOptimizer._adjust_signs_df()` on different types of inputs.
+    """
+    df_scores_input = pandas.DataFrame({
+        'latency': [88.88, 66.66, 99.99, None],
+        'throughput': [111, 222, 333, None],
+    })
+
+    df_scores_output = pandas.DataFrame({
+        'latency': [88.88, 66.66, 99.99, float("NaN")],
+        'throughput': [-111, -222, -333, float("NaN")],
+    })
+
+    # Make sure we adjust the signs for minimization.
+    df_scores = mlos_core_optimizer._adjust_signs_df(df_scores_input)
+    assert df_scores.equals(df_scores_output)
+
+    # Check that the same operation works for string inputs.
+    df_scores = mlos_core_optimizer._adjust_signs_df(df_scores_input.astype(str))
+    assert df_scores.equals(df_scores_output)