A small Python library for replaying real human mouse trajectories from the SapiMouse dataset onto a Playwright page. Designed for ML-based bot-detection research, behavioral biometrics prototyping, and replay-based test fixtures. Public API: load_all_segments, pick_segments, affine_warp, upsample, replay, replay_random, download_sapimouse. - src/ layout with hatchling build backend - 23 pytest tests (10 transform unit + 13 integration) - MIT license, PEP 561 py.typed marker - python -m human_mouse download for one-shot dataset fetch - examples/cloakbrowser_demo.py demonstrates end-to-end use with CloakBrowser Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
155 lines
4.7 KiB
Python
155 lines
4.7 KiB
Python
"""Integration tests: real SapiMouse loading, filtering, caching, error paths.
|
|
|
|
These tests assume the SapiMouse dataset is present at the standard project
|
|
location (sapimouse_data/sapimouse/). They will skip otherwise — so the
|
|
core unit tests still pass on a fresh clone.
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
import math
|
|
from pathlib import Path
|
|
|
|
import pytest
|
|
|
|
import human_mouse as hm
|
|
from human_mouse import _replay as replay_mod
|
|
|
|
REPO_ROOT = Path(__file__).resolve().parents[1]
|
|
DATA_ROOT = REPO_ROOT / "sapimouse_data" / "sapimouse"
|
|
|
|
pytestmark = pytest.mark.skipif(
|
|
not DATA_ROOT.exists(),
|
|
reason="SapiMouse dataset not available at the default location",
|
|
)
|
|
|
|
|
|
# --------------------------------------------------------- happy-path loading
|
|
|
|
|
|
def test_load_all_segments_finds_segments():
|
|
segs = hm.load_all_segments(DATA_ROOT)
|
|
assert len(segs) > 1000, f"expected thousands, got {len(segs)}"
|
|
assert all(len(s.points) >= 50 for s in segs)
|
|
|
|
|
|
def test_segment_properties_consistent():
|
|
segs = hm.load_all_segments(DATA_ROOT)
|
|
s = segs[0]
|
|
sx, sy = s.start
|
|
ex, ey = s.end
|
|
assert s.straight_distance == pytest.approx(math.hypot(ex - sx, ey - sy))
|
|
assert s.path_length >= s.straight_distance - 1e-6
|
|
assert s.path_ratio >= 1.0 - 1e-6
|
|
assert s.duration_ms >= 0
|
|
|
|
|
|
# -------------------------------------------------------------- pick_segments
|
|
|
|
|
|
def test_pick_segments_basic():
|
|
segs = hm.load_all_segments(DATA_ROOT)
|
|
picked = hm.pick_segments(segs, n=3, target_distance=500, seed=42)
|
|
assert len(picked) == 3
|
|
assert all(0.25 * 500 < p.straight_distance < 1.75 * 500 for p in picked)
|
|
|
|
|
|
def test_pick_segments_distinct_sessions():
|
|
segs = hm.load_all_segments(DATA_ROOT)
|
|
picked = hm.pick_segments(segs, n=5, target_distance=800,
|
|
distinct_sessions=True, seed=1)
|
|
keys = {(p.user, p.session) for p in picked}
|
|
assert len(keys) == 5
|
|
|
|
|
|
def test_pick_segments_seed_is_deterministic():
|
|
segs = hm.load_all_segments(DATA_ROOT)
|
|
a = hm.pick_segments(segs, n=3, target_distance=800, seed=7)
|
|
b = hm.pick_segments(segs, n=3, target_distance=800, seed=7)
|
|
assert [(p.user, p.session) for p in a] == [(p.user, p.session) for p in b]
|
|
|
|
|
|
def test_pick_segments_path_ratio_filter():
|
|
segs = hm.load_all_segments(DATA_ROOT)
|
|
picked = hm.pick_segments(segs, n=10, target_distance=800,
|
|
max_path_ratio=1.3, seed=1)
|
|
assert all(p.path_ratio <= 1.3 for p in picked)
|
|
|
|
|
|
# ------------------------------------------------------------------ errors
|
|
|
|
|
|
def test_load_missing_directory_raises():
|
|
with pytest.raises(FileNotFoundError, match="SapiMouse"):
|
|
hm.load_all_segments("/nonexistent/path")
|
|
|
|
|
|
def test_pick_segments_no_match_raises():
|
|
segs = hm.load_all_segments(DATA_ROOT)
|
|
with pytest.raises(ValueError, match="No segments matched"):
|
|
hm.pick_segments(segs, n=1, target_distance=999_999,
|
|
distance_tolerance=0.01)
|
|
|
|
|
|
def test_pick_segments_not_enough_distinct_raises():
|
|
segs = hm.load_all_segments(DATA_ROOT)
|
|
with pytest.raises(ValueError, match="distinct"):
|
|
hm.pick_segments(segs, n=10_000, target_distance=800)
|
|
|
|
|
|
# ------------------------------------------------------------------- cache
|
|
|
|
|
|
def test_replay_cache_reuses_same_data_root(monkeypatch):
|
|
replay_mod._SEGMENT_CACHE.clear()
|
|
calls = {"n": 0}
|
|
original = hm.load_all_segments
|
|
|
|
def counting(*a, **kw):
|
|
calls["n"] += 1
|
|
return original(*a, **kw)
|
|
|
|
monkeypatch.setattr(replay_mod, "load_all_segments", counting)
|
|
_ = replay_mod._cached_segments(DATA_ROOT)
|
|
_ = replay_mod._cached_segments(DATA_ROOT)
|
|
_ = replay_mod._cached_segments(DATA_ROOT)
|
|
assert calls["n"] == 1
|
|
|
|
|
|
# ------------------------------------------------------ end-to-end replay_random
|
|
|
|
|
|
class _FakeMouse:
|
|
def __init__(self) -> None:
|
|
self.moves: list[tuple[int, int]] = []
|
|
|
|
def move(self, x: int, y: int) -> None:
|
|
self.moves.append((x, y))
|
|
|
|
|
|
class _FakePage:
|
|
"""A Page-like duck that satisfies the _PageLike Protocol."""
|
|
def __init__(self) -> None:
|
|
self.mouse = _FakeMouse()
|
|
|
|
|
|
def test_replay_random_drives_page_endpoints():
|
|
page = _FakePage()
|
|
seg = hm.replay_random(
|
|
page, start=(100, 100), end=(900, 500),
|
|
data_root=DATA_ROOT, density=2, speed=10_000.0, seed=1,
|
|
)
|
|
assert isinstance(seg, hm.Segment)
|
|
assert len(page.mouse.moves) >= 50
|
|
assert page.mouse.moves[0] == (100, 100)
|
|
assert page.mouse.moves[-1] == (900, 500)
|
|
|
|
|
|
def test_replay_random_returns_filtered_segment():
|
|
page = _FakePage()
|
|
seg = hm.replay_random(
|
|
page, start=(0, 0), end=(800, 0),
|
|
data_root=DATA_ROOT, density=1, speed=10_000.0, seed=2,
|
|
max_path_ratio=1.5,
|
|
)
|
|
assert seg.path_ratio <= 1.5
|