From 73b2e0547cb44e55818587e133645e5ffa28d06f Mon Sep 17 00:00:00 2001
From: Daniel Gaskins <danielgaskins99@gmail.com>
Date: Thu, 20 Aug 2026 13:24:06 -0700
Subject: [PATCH] fix: make NumPy object arrays JSON-safe

Signed-off-by: Daniel Gaskins <danielgaskins99@gmail.com>
---
 aisteer360/evaluation/utils/data_utils.py |  6 ++---
 tests/core/test_benchmark.py              | 30 +++++++++++++++++++++--
 tests/core/test_evaluation_utils.py       | 19 ++++++++++++--
 3 files changed, 48 insertions(+), 7 deletions(-)

diff --git a/aisteer360/evaluation/utils/data_utils.py b/aisteer360/evaluation/utils/data_utils.py
index 5b329cfb..0f1277c5 100644
--- a/aisteer360/evaluation/utils/data_utils.py
+++ b/aisteer360/evaluation/utils/data_utils.py
@@ -13,7 +13,7 @@ def to_jsonable(obj: Any) -> Any:
     - Path: str(path)
     - mappings: recurse, stringify keys
     - sequences: recurse on elements
-    - numpy scalars/arrays: convert to Python / list
+    - numpy scalars/arrays: convert to Python / list, then recurse
     - everything else: repr(obj)
     """
     from pathlib import Path as _Path
@@ -28,7 +28,7 @@ def to_jsonable(obj: Any) -> Any:
         return obj.item()
 
     if isinstance(obj, np.ndarray):
-        return obj.tolist()
+        return to_jsonable(obj.tolist())
 
     if isinstance(obj, Mapping):
         return {str(k): to_jsonable(v) for k, v in obj.items()}
@@ -38,7 +38,7 @@ def to_jsonable(obj: Any) -> Any:
 
     if callable(obj):
         return f"callable:{getattr(obj, '__qualname__', type(obj).__name__)}"
-    
+
     return repr(obj)
 
 
diff --git a/tests/core/test_benchmark.py b/tests/core/test_benchmark.py
index e94ce82c..9d876b8b 100644
--- a/tests/core/test_benchmark.py
+++ b/tests/core/test_benchmark.py
@@ -16,8 +16,10 @@
 test are the package implementations.
 """
 import json
+from pathlib import Path
 from unittest.mock import MagicMock
 
+import numpy as np
 import pytest
 
 from aisteer360.algorithms.core.specs import ControlSpec
@@ -118,8 +120,6 @@ def test_num_trials_coerced_to_int(self, sample_evaluation_data):
         assert isinstance(benchmark.num_trials, int)
 
     def test_save_dir_coerced_to_path(self, sample_evaluation_data, tmp_path):
-        from pathlib import Path
-
         benchmark = Benchmark(
             use_case=_make_use_case(sample_evaluation_data),
             base_model_name_or_path="test-model",
@@ -439,6 +439,32 @@ def test_mixed_spec_and_concrete_raises(self, sample_evaluation_data, mock_base_
 class TestBenchmarkCheckpointing:
     """Tests for incremental checkpointing and resume."""
 
+    def test_checkpoint_converts_nested_numpy_objects(self, sample_evaluation_data, tmp_path):
+        benchmark = Benchmark(
+            use_case=_make_use_case(sample_evaluation_data),
+            base_model_name_or_path="test-model",
+            steering_pipelines={"baseline": []},
+            save_dir=tmp_path,
+        )
+        profiles = {
+            "baseline": [
+                {
+                    "params": np.array(
+                        [Path("models/base"), {"layers": {1, 2}}],
+                        dtype=object,
+                    )
+                }
+            ]
+        }
+
+        benchmark._save_checkpoint(profiles)
+
+        with (tmp_path / "checkpoint.json").open(encoding="utf-8") as handle:
+            saved = json.load(handle)
+        saved_params = saved["baseline"][0]["params"]
+        assert saved_params[0] == "models/base"
+        assert sorted(saved_params[1]["layers"]) == [1, 2]
+
     def test_checkpoint_written(self, sample_evaluation_data, mock_base_model, tmp_path):
         use_case = _make_use_case(sample_evaluation_data)
         benchmark = Benchmark(
diff --git a/tests/core/test_evaluation_utils.py b/tests/core/test_evaluation_utils.py
index c83954c5..ee64a97a 100644
--- a/tests/core/test_evaluation_utils.py
+++ b/tests/core/test_evaluation_utils.py
@@ -7,6 +7,9 @@
 - Visualization utilities (plot functions)
 """
 
+import json
+from pathlib import Path
+
 import numpy as np
 import pandas as pd
 import pytest
@@ -149,8 +152,6 @@ def test_primitive_passthrough(self):
 
     def test_path_to_string(self):
         """Test that Path objects become strings."""
-        from pathlib import Path
-
         result = to_jsonable(Path("/some/path"))
         assert result == "/some/path"
         assert isinstance(result, str)
@@ -168,6 +169,20 @@ def test_numpy_array(self):
         assert result == [1, 2, 3]
         assert isinstance(result, list)
 
+    def test_numpy_object_array_recurses(self):
+        """Test nested objects in numpy arrays become JSON-safe."""
+        arr = np.array(
+            [Path("models/base"), {"layers": {1, 2}}],
+            dtype=object,
+        )
+
+        result = to_jsonable(arr)
+        encoded = json.dumps(result)
+
+        assert result[0] == "models/base"
+        assert sorted(result[1]["layers"]) == [1, 2]
+        assert json.loads(encoded) == result
+
     def test_nested_dict(self):
         """Test nested dict conversion."""
         data = {"a": np.float64(1.0), "b": {"c": np.array([1, 2])}}