74 lines
3.1 KiB
Python
74 lines
3.1 KiB
Python
"""TensorBoard and measurement checks using CPU tensors, never a simulator."""
|
|
|
|
import csv
|
|
from types import SimpleNamespace
|
|
|
|
import pytest
|
|
|
|
torch = pytest.importorskip("torch")
|
|
pytest.importorskip("tensorboard")
|
|
|
|
from torch.utils.tensorboard import SummaryWriter # noqa: E402
|
|
|
|
from universal_humanoid.artifacts import read_json # noqa: E402
|
|
from universal_humanoid.curves import export_curves, read_curves # noqa: E402
|
|
from universal_humanoid.measurement import JointRecorder # noqa: E402
|
|
|
|
|
|
def test_tensorboard_round_trip_and_missing_tags(tmp_path):
|
|
log = tmp_path / "log"
|
|
with SummaryWriter(str(log)) as writer:
|
|
for i in range(3):
|
|
writer.add_scalar("Episode_Reward/track_linear_velocity", i / 4, i)
|
|
writer.add_scalar("Policy/mean_std", 1 - i / 10, i)
|
|
writer.add_scalar("Train/mean_episode_length", 10 + i, i)
|
|
writer.add_scalar("Train/mean_episode_length/time", 10, 1)
|
|
writer.add_scalar("Train/mean_episode_length/time", 20, 2)
|
|
writer.add_scalar("Loss/bad", float("nan"), 0)
|
|
result = export_curves(log, tmp_path / "curves")
|
|
assert "tracking" not in result["missing_categories"]
|
|
assert result["curves"]["Train/mean_episode_length"]["last"] == 12
|
|
assert (tmp_path / "curves/curves.svg").is_file()
|
|
assert "Train/mean_episode_length/time" not in result["curves"]
|
|
wall_time = read_curves(log, match="/time$")["series"]["Train/mean_episode_length/time"]
|
|
assert wall_time["step_axis"] == "elapsed_seconds"
|
|
assert wall_time["summary"]["tail_slope_per_second"] == 10
|
|
with pytest.raises(ValueError, match="regex"):
|
|
read_curves(log, match="[")
|
|
assert read_curves(log, match="Loss/bad")["series"]["Loss/bad"]["nonfinite_points"] == 1
|
|
with pytest.raises(FileExistsError):
|
|
export_curves(log, tmp_path / "curves")
|
|
with pytest.raises(FileNotFoundError):
|
|
read_curves(tmp_path / "missing")
|
|
|
|
|
|
def test_joint_csv_uses_named_order_and_projected_torque(tmp_path):
|
|
robot = SimpleNamespace(
|
|
joint_names=["a", "b"],
|
|
data=SimpleNamespace(
|
|
joint_pos=torch.tensor([[1.0, 2.0]]),
|
|
joint_vel=torch.tensor([[3.0, 4.0]]),
|
|
qfrc_actuator=torch.tensor([[5.0, 6.0]]),
|
|
joint_pos_target=torch.tensor([[7.0, 8.0]]),
|
|
),
|
|
)
|
|
term = SimpleNamespace(target_names=["b", "a"])
|
|
raw = SimpleNamespace(
|
|
scene={"robot": robot},
|
|
step_dt=0.02,
|
|
action_manager=SimpleNamespace(get_term=lambda _: term),
|
|
)
|
|
path = tmp_path / "measure.csv"
|
|
recorder = JointRecorder(path, SimpleNamespace(unwrapped=raw, clip_actions=5))
|
|
recorder.record(0, 0.02, [0], [1], [True], [False], torch.tensor([[10.0, -10.0]]))
|
|
recorder.close(True)
|
|
with path.open() as stream:
|
|
rows = list(csv.DictReader(stream))
|
|
assert [r["joint"] for r in rows] == ["b", "a"]
|
|
assert float(rows[0]["position_rad"]) == 2
|
|
assert float(rows[0]["actuator_joint_torque_nm"]) == 6
|
|
assert float(rows[0]["applied_action"]) == 5
|
|
assert float(rows[0]["target_rad"]) == 8
|
|
assert rows[0]["terminated"] == "1"
|
|
assert read_json(path.with_suffix(".csv.json"))["rows"] == 2
|