code/VLAarchtests2_code/VLAarchtests/tests/test_adapter_dataset_alignment.py

from pathlib import Path

from omegaconf import OmegaConf

import train.run_experiment as run_experiment


def test_adapter_dataset_rebuilds_when_existing_bundle_lacks_proposal_targets(monkeypatch, tmp_path):
    dataset_path = tmp_path / "proxy_train.pt"
    dataset_path.write_text("stub", encoding="utf-8")
    builder = object()
    captured: dict[str, object] = {}

    monkeypatch.setattr(
        run_experiment,
        "load_teacher_dataset",
        lambda path: {"samples": [{"task_name": "bag"}]},
    )

    def _fake_collect_teacher_dataset(**kwargs):
        captured.update(kwargs)
        return {
            "samples": [
                {
                    "task_name": "bag",
                    "proposal_target_action_chunks": 1,
                    "proposal_target_retrieval_success": 1,
                    "proposal_target_risk": 1,
                    "proposal_target_utility": 1,
                }
            ]
        }

    monkeypatch.setattr(run_experiment, "collect_teacher_dataset", _fake_collect_teacher_dataset)
    monkeypatch.setattr(run_experiment, "save_teacher_dataset", lambda path, bundle: Path(path))

    data_cfg = OmegaConf.create(
        {
            "proxies": ["bag_proxy"],
            "resolution": 16,
            "seed": 17,
            "chunk_horizon": 2,
            "rollout_horizon": 2,
            "history_steps": 1,
            "planner_candidates": 2,
            "dataset_version": "proxy_test",
            "train_episodes_per_proxy": 1,
            "train_dataset_path": str(dataset_path),
            "rebuild_dataset": False,
        }
    )

    bundle = run_experiment._build_dataset_from_config(
        data_cfg,
        "train",
        proposal_target_builder=builder,
        require_proposal_targets=True,
    )

    assert run_experiment._bundle_has_proposal_targets(bundle)
    assert captured["proposal_target_builder"] is builder


def test_adapter_dataset_missing_proposal_targets_raises_without_builder(monkeypatch, tmp_path):
    dataset_path = tmp_path / "proxy_train.pt"
    dataset_path.write_text("stub", encoding="utf-8")
    monkeypatch.setattr(
        run_experiment,
        "load_teacher_dataset",
        lambda path: {"samples": [{"task_name": "bag"}]},
    )

    data_cfg = OmegaConf.create(
        {
            "proxies": ["bag_proxy"],
            "resolution": 16,
            "seed": 17,
            "chunk_horizon": 2,
            "rollout_horizon": 2,
            "history_steps": 1,
            "planner_candidates": 2,
            "dataset_version": "proxy_test",
            "train_episodes_per_proxy": 1,
            "train_dataset_path": str(dataset_path),
            "rebuild_dataset": False,
        }
    )

    try:
        run_experiment._build_dataset_from_config(
            data_cfg,
            "train",
            proposal_target_builder=None,
            require_proposal_targets=True,
        )
    except RuntimeError as exc:
        assert "proposal-aligned targets" in str(exc)
    else:
        raise AssertionError("Expected a RuntimeError for unaligned adapter dataset.")


def test_adapter_dataset_rebuilds_when_transition_rollout_targets_are_missing(monkeypatch, tmp_path):
    dataset_path = tmp_path / "proxy_train.pt"
    dataset_path.write_text("stub", encoding="utf-8")
    builder = object()
    captured: dict[str, object] = {}

    monkeypatch.setattr(
        run_experiment,
        "load_teacher_dataset",
        lambda path: {
            "samples": [
                {
                    "task_name": "bag",
                    "proposal_target_action_chunks": 1,
                    "proposal_target_retrieval_success": 1,
                    "proposal_target_risk": 1,
                    "proposal_target_utility": 1,
                }
            ]
        },
    )

    def _fake_collect_teacher_dataset(**kwargs):
        captured.update(kwargs)
        return {
            "samples": [
                {
                    "task_name": "bag",
                    "proposal_target_action_chunks": 1,
                    "proposal_target_retrieval_success": 1,
                    "proposal_target_risk": 1,
                    "proposal_target_utility": 1,
                    "proposal_target_rollout_support_mode": 1,
                    "proposal_target_rollout_corridor_feasible": 1,
                    "proposal_target_rollout_persistence_horizon": 1,
                    "proposal_target_rollout_disturbance_cost": 1,
                }
            ]
        }

    monkeypatch.setattr(run_experiment, "collect_teacher_dataset", _fake_collect_teacher_dataset)
    monkeypatch.setattr(run_experiment, "save_teacher_dataset", lambda path, bundle: Path(path))

    data_cfg = OmegaConf.create(
        {
            "proxies": ["bag_proxy"],
            "resolution": 16,
            "seed": 17,
            "chunk_horizon": 2,
            "rollout_horizon": 2,
            "history_steps": 1,
            "planner_candidates": 2,
            "dataset_version": "proxy_test",
            "train_episodes_per_proxy": 1,
            "train_dataset_path": str(dataset_path),
            "rebuild_dataset": False,
        }
    )

    bundle = run_experiment._build_dataset_from_config(
        data_cfg,
        "train",
        proposal_target_builder=builder,
        require_proposal_targets=True,
        require_proposal_rollout_targets=True,
    )

    assert run_experiment._bundle_has_proposal_rollout_targets(bundle)
    assert captured["proposal_target_builder"] is builder