Test replay of batch mode

Author: klieret

sweagent/environment/utils.py

@@ -519,6 +519,9 @@ def instance_from_dict(instances):
         return ib.build()
 
     def postproc_instance_list(instances):
+        if isinstance(instances, dict):
+            msg = "Expected a list of instances, got a dictionary."
+            raise ValueError(msg)
         return [instance_from_dict(x) for x in instances]
 
 

tests/conftest.py

@@ -26,6 +26,13 @@ def test_trajectories_path(test_data_path) -> Path:
     return p
 
 
+@fixture
+def test_data_sources_path(test_data_path) -> Path:
+    p = test_data_path / "data_sources"
+    assert p.is_dir()
+    return p
+
+
 @fixture
 def test_trajectory(test_trajectories_path):
     traj_dir = test_trajectories_path / "gpt4__klieret__swe-agent-test-repo__default_from_url__t-0.00__p-0.95__c-3.00__install-1" / "klieret__swe-agent-test-repo-i1.traj"

tests/test_data/data_sources/swe-bench-dev-easy_first_only.json

@@ -0,0 +1 @@
+{"model_name_or_path": "gpt4__swe-bench-dev-easy_first_only__default__t-0.00__p-0.95__c-3.00__install-1", "instance_id": "pydicom__pydicom-1458", "model_patch": "\ndiff --git a/pydicom/pixel_data_handlers/numpy_handler.py b/pydicom/pixel_data_handlers/numpy_handler.py\nindex 8e8d319ae..6a1221b72 100644\n--- a/pydicom/pixel_data_handlers/numpy_handler.py\n+++ b/pydicom/pixel_data_handlers/numpy_handler.py\n@@ -285,9 +285,10 @@ def get_pixeldata(ds: \"Dataset\", read_only: bool = False) -> \"np.ndarray\":\n         )\n \n     required_elements = [\n-        'BitsAllocated', 'Rows', 'Columns', 'PixelRepresentation',\n-        'SamplesPerPixel', 'PhotometricInterpretation'\n+        'BitsAllocated', 'Rows', 'Columns', 'SamplesPerPixel', 'PhotometricInterpretation'\n     ]\n+    if 'PixelData' in ds:\n+        required_elements.append('PixelRepresentation')\n     missing = [elem for elem in required_elements if elem not in ds]\n     if missing:\n         raise AttributeError(\n"}

tests/test_data/trajectories/gpt4__swe-bench-dev-easy_first_only__default__t-0.00__p-0.95__c-3.00__install-1/all_preds.jsonl

@@ -0,0 +1,18 @@
+
+diff --git a/pydicom/pixel_data_handlers/numpy_handler.py b/pydicom/pixel_data_handlers/numpy_handler.py
+index 8e8d319ae..6a1221b72 100644
+--- a/pydicom/pixel_data_handlers/numpy_handler.py
++++ b/pydicom/pixel_data_handlers/numpy_handler.py
+@@ -285,9 +285,10 @@ def get_pixeldata(ds: "Dataset", read_only: bool = False) -> "np.ndarray":
+         )
+ 
+     required_elements = [
+-        'BitsAllocated', 'Rows', 'Columns', 'PixelRepresentation',
+-        'SamplesPerPixel', 'PhotometricInterpretation'
++        'BitsAllocated', 'Rows', 'Columns', 'SamplesPerPixel', 'PhotometricInterpretation'
+     ]
++    if 'PixelData' in ds:
++        required_elements.append('PixelRepresentation')
+     missing = [elem for elem in required_elements if elem not in ds]
+     if missing:
+         raise AttributeError(

tests/test_data/trajectories/gpt4__swe-bench-dev-easy_first_only__default__t-0.00__p-0.95__c-3.00__install-1/args.yaml

@@ -48,6 +48,27 @@ def test_model_replay_github_repo(swe_agent_test_repo_traj, problem_statement_so
     main(**vars(args), forward_args=remaining_args)
 
 
+@pytest.mark.slow
+def test_model_replay_from_json(test_trajectories_path, test_data_sources_path):
+    traj_path = test_trajectories_path / "gpt4__swe-bench-dev-easy_first_only__default__t-0.00__p-0.95__c-3.00__install-1" / "pydicom__pydicom-1458.traj" 
+    assert traj_path.is_file()
+    data_path = test_data_sources_path / "swe-bench-dev-easy_first_only.json"
+    assert data_path.is_file()
+    args = [
+        "--traj_path",
+        str(traj_path),
+        "--data_path",
+        str(data_path),
+        "--config_file",
+        "config/default.yaml",
+        "--raise_exceptions",
+    ]
+    args, remaining_args = get_args(args)
+    main(**vars(args), forward_args=remaining_args)
+
+
+
+
 def test_run_cli_help():
     args = [
         "python",