tests/test_evaluation_tools.py

import json
import sys
import tempfile
import unittest
from itertools import chain, repeat
from pathlib import Path
from unittest import mock

sys.path.insert(0, str(Path(__file__).resolve().parents[1]))

from evaluation import context_eval_baselines
from evaluation.tools import to_pred


class ToPredTests(unittest.TestCase):
    def test_task_is_required(self):
        parser = to_pred.build_parser()

        with self.assertRaises(SystemExit):
            parser.parse_args(["--input", "pred.json", "--output", "out.json"])

    def test_convert_old_predictions_to_semantics_fields(self):
        with tempfile.TemporaryDirectory() as tmpdir:
            tmp = Path(tmpdir)
            pred_path = tmp / "pred.json"
            question_path = tmp / "questions.jsonl"
            output_path = tmp / "out.json"

            pred_path.write_text(
                json.dumps(
                    {
                        "q1": {
                            "oovd_result": {
                                "button": [
                                    {
                                        "bbox_pixels": [1, 2, 3, 4],
                                        "probability": 0.7,
                                    }
                                ]
                            }
                        }
                    }
                )
            )
            question_path.write_text(
                json.dumps({"question_id": "q1", "image": "1026760_11.jpg"}) + "\n"
            )

            to_pred.main(
                [
                    "--task",
                    "semantics",
                    "--input",
                    str(pred_path),
                    "--questions",
                    str(question_path),
                    "--output",
                    str(output_path),
                ]
            )

            converted = json.loads(output_path.read_text())
            self.assertEqual(
                converted,
                [
                    {
                        "image_id": 1026760011,
                        "category_id": "button",
                        "bbox": [1, 2, 3, 4],
                        "score": 0.7,
                    }
                ],
            )

    def test_all_outputs_do_not_overwrite(self):
        with tempfile.TemporaryDirectory() as tmpdir:
            tmp = Path(tmpdir)
            pred_path = tmp / "pred.json"
            output_path = tmp / "converted.json"

            pred_path.write_text(
                json.dumps(
                    [
                        {
                            "image_id": 1,
                            "category_id": "trigger",
                            "bbox": [1, 2, 3, 4],
                            "score": 0.5,
                        }
                    ]
                )
            )

            to_pred.main(
                [
                    "--task",
                    "all",
                    "--input",
                    str(pred_path),
                    "--output",
                    str(output_path),
                ]
            )

            self.assertFalse(output_path.exists())
            for task in to_pred.TASKS:
                self.assertTrue((tmp / f"converted_{task}.json").exists())

            interactable = json.loads((tmp / "converted_interactable.json").read_text())
            interaction = json.loads((tmp / "converted_interaction.json").read_text())
            semantics = json.loads((tmp / "converted_semantics.json").read_text())
            self.assertEqual(interactable[0]["category_id"], 1)
            self.assertEqual(interaction[0]["category_id"], "trigger")
            self.assertEqual(semantics[0]["category_id"], "trigger")


class ContextEvalBaselineTests(unittest.TestCase):
    def test_validate_unique_methods_rejects_duplicates(self):
        with self.assertRaises(ValueError):
            context_eval_baselines.validate_unique_methods(["Seed-E2E", "Seed-E2E"])

    def test_main_uses_subprocess_check_true(self):
        args = context_eval_baselines.build_parser().parse_args([])
        methods = ["CenterNet2"]

        with mock.patch.object(
            context_eval_baselines, "METHODS", methods
        ), mock.patch.object(
            context_eval_baselines, "LLM_METHODS", []
        ), mock.patch.object(
            context_eval_baselines.os.path,
            "exists",
            side_effect=chain([True, False], repeat(False)),
        ), mock.patch.object(
            context_eval_baselines.subprocess, "run"
        ) as run:
            context_eval_baselines.main(args)

        self.assertTrue(run.called)
        self.assertTrue(run.call_args.kwargs["check"])
        self.assertIsInstance(run.call_args.args[0], list)


if __name__ == "__main__":
    unittest.main()