File size: 15,197 Bytes
10ef792 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 | """Small synthetic tests; run only on allocated compute, never login nodes."""
from __future__ import annotations
import hashlib
import json
import pickle
import sys
import numpy as np
import pytest
from wisp_release import list_methods, load_checkpoint, predict_checkpoint
from wisp_release.checkpoints import require_sequence_metadata
from wisp_release.cli import main
from wisp_release.data import as_har_data, read_npz
from wisp_release.methods import build_family
from wisp_release.selection import choose_validation_winner, fit_selector
from wisp_release.scoring import score
class RecordingModel:
def __init__(self, sequence: bool = False):
self.smoother_ = object() if sequence else None
self.n_jobs = 99
self.called = False
def predict(self, X, *, groups=None, time_index=None):
self.called = True
self.last_groups = groups
self.last_time_index = time_index
return np.zeros(len(X), dtype=np.int64)
class SelectorFixture:
fit_records = []
def __init__(self, family):
self.family = family
self.smoother_ = None
self.n_jobs = 1
def fit(self, X, y, *, groups, time_index, n_classes):
self.fit_records.append((self.family, len(X), n_classes))
self.classes_ = np.arange(n_classes)
return self
def predict(self, X, *, groups=None, time_index=None):
return X[:, 0, 0].astype(np.int64) if self.family == "wisp_rc" else np.zeros(len(X), dtype=np.int64)
def test_full_series_has_only_paper_methods():
rows = list_methods()
assert len(rows) == 11
assert len({row["method_id"] for row in rows}) == 11
assert all(row["paper_name"].startswith("WISP-") for row in rows)
assert sum(row["family_fit"] for row in rows) == 7
assert sum(row["search"] for row in rows) == 2
assert all("baseline" not in row["method_id"] for row in rows)
def test_pickle_load_requires_explicit_trust_before_open(tmp_path):
with pytest.raises(ValueError, match="arbitrary code"):
load_checkpoint(tmp_path / "does_not_exist.pkl")
def test_sha256_verified_before_unpickling(tmp_path, monkeypatch):
path = tmp_path / "model.pkl"
path.write_bytes(pickle.dumps(RecordingModel()))
with pytest.raises(ValueError, match="SHA256 mismatch"):
load_checkpoint(path, trust_pickle=True, expected_sha256="0" * 64)
with pytest.raises(ValueError, match="hexadecimal"):
load_checkpoint(path, trust_pickle=True, expected_sha256="not-a-digest")
digest = hashlib.sha256(path.read_bytes()).hexdigest()
assert isinstance(load_checkpoint(path, trust_pickle=True, expected_sha256=digest), RecordingModel)
def test_legacy_class_namespace_without_global_alias(tmp_path):
from wisp.cpu.algorithms import HarSculptForestHMM
model = HarSculptForestHMM(seed=123, n_jobs=1)
original = pickle.dumps(model, protocol=2)
legacy = original.replace(b"cwisp.cpu.algorithms\n", b"ctsevolve.cpu.algorithms\n")
assert original != legacy
path = tmp_path / "legacy.pkl"
path.write_bytes(legacy)
before_modules = {key for key in sys.modules if key == "tsevolve" or key.startswith("tsevolve.")}
loaded = load_checkpoint(path, trust_pickle=True)
assert isinstance(loaded, HarSculptForestHMM)
assert loaded.seed == 123
assert path.read_bytes() == legacy
after_modules = {key for key in sys.modules if key == "tsevolve" or key.startswith("tsevolve.")}
assert after_modules == before_modules
@pytest.mark.parametrize("missing", ["subject", "time_index", "both"])
def test_sequence_decoder_cannot_silently_be_disabled(missing):
model = RecordingModel(sequence=True)
subject = None if missing in ("subject", "both") else np.array([1, 1, 2])
time_index = None if missing in ("time_index", "both") else np.arange(3)
with pytest.raises(ValueError, match="both subject"):
predict_checkpoint(model, np.ones((3, 8, 3)), subject=subject, time_index=time_index)
assert not model.called
def test_sequence_prediction_retains_groups_times_and_limits_jobs():
model = RecordingModel(sequence=True)
subject = np.array(["recording_a", "recording_a", "recording_b"])
time = np.array([0, 1, 0])
pred = predict_checkpoint(model, np.ones((3, 8, 3)), subject=subject, time_index=time,
metadata={"input_shape": [8, 3], "label_names": ["rest", "walk"]})
np.testing.assert_array_equal(pred, [0, 0, 0])
np.testing.assert_array_equal(model.last_groups, subject)
np.testing.assert_array_equal(model.last_time_index, time)
assert model.n_jobs == 1
def test_shape_and_finite_value_guards():
model = RecordingModel()
with pytest.raises(ValueError, match="time/channel"):
predict_checkpoint(model, np.ones((3, 8, 3)), metadata={"input_shape": [16, 3]})
with pytest.raises(ValueError, match="NaN"):
predict_checkpoint(model, np.full((3, 8, 3), np.nan))
with pytest.raises(ValueError, match="shape"):
predict_checkpoint(model, np.ones((3, 8)))
with pytest.raises(ValueError, match="one-dimensional"):
require_sequence_metadata(model, [[1, 1, 1]], None, n=3)
def test_short_vocabulary_rejected_even_when_missing_class_not_predicted():
model = RecordingModel()
model.classes_ = np.arange(3)
with pytest.raises(ValueError, match="preserve the complete fitted vocabulary"):
predict_checkpoint(model, np.ones((3, 8, 3)), metadata={"label_names": ["rest", "walk"]})
assert model.called is False
@pytest.mark.parametrize("classes", [np.array([0, 2]), np.array([1, 0]), np.array([0, 0]),
np.array([0.0, 1.0]), np.array([], dtype=np.int64)])
def test_nonstandard_checkpoint_class_axis_rejected_without_remapping(classes):
model = RecordingModel()
model.classes_ = classes
with pytest.raises(ValueError, match="contiguous encoded global axis"):
predict_checkpoint(model, np.ones((3, 8, 3)))
assert model.called is False
np.testing.assert_array_equal(model.classes_, classes)
def test_npz_and_checkpoint_sidecar_vocab_conflict(tmp_path):
model = RecordingModel()
model.classes_ = np.arange(3)
checkpoint = tmp_path / "model.pkl"
checkpoint.write_bytes(pickle.dumps(model))
(tmp_path / "checkpoint.json").write_text(json.dumps({"input_shape": [8, 3],
"label_names": ["rest", "walk", "run"]}))
data = tmp_path / "test.npz"
np.savez(data, X=np.ones((3, 8, 3)), y=np.zeros(3, dtype=np.int64),
label_names=np.array(["rest", "run", "walk"]))
with pytest.raises(SystemExit) as error:
main(["evaluate", "--checkpoint", str(checkpoint), "--input", str(data),
"--output", str(tmp_path / "metrics.json"), "--trust-pickle"])
assert error.value.code == 2
assert not (tmp_path / "metrics.json").exists()
def test_numeric_npz_no_embedded_pickle(tmp_path):
path = tmp_path / "safe.npz"
np.savez(path, X=np.ones((4, 8, 3)), y=np.array([0, 2, 2, 0]),
subject=np.array(["a", "a", "b", "b"]), time_index=np.array([0, 1, 0, 1]),
label_names=np.array(["rest", "absent-in-fold", "walk"]))
data = read_npz(path, require_y=True)
dataset = as_har_data(data)
assert dataset.n_classes == 3
assert dataset.label_names == ["rest", "absent-in-fold", "walk"]
unsafe = tmp_path / "unsafe.npz"
np.savez(unsafe, X=np.ones((4, 8, 3)), y=np.array([0, 2, 2, 0]),
subject=np.asarray([{}, {}, {}, {}], dtype=object))
with pytest.raises(ValueError, match="object arrays"):
read_npz(unsafe)
def test_global_class_axis_requires_explicit_vocabulary():
data = {"X": np.ones((4, 8, 3)), "y": np.array([0, 2, 2, 0])}
with pytest.raises(ValueError, match="global label_names"):
as_har_data(data)
assert as_har_data(data, n_classes=4).n_classes == 4
def test_cli_training_is_plan_only_by_default(capsys, tmp_path):
assert main(["family-fit", "--method", "wisp_cis", "--input", "not-present.npz",
"--output", str(tmp_path / "model.pkl"), "--seed", "42", "--n-jobs", "1"]) == 0
result = json.loads(capsys.readouterr().out)
assert result["execute"] is False
assert not (tmp_path / "model.pkl").exists()
assert main(["search", "--method", "wisp_random", "--train", "train.npz", "--valid", "valid.npz",
"--test", "test.npz", "--output", str(tmp_path / "search"), "--population", "2",
"--generations", "1", "--max-evaluations", "4", "--seed", "42", "--n-jobs", "1"]) == 0
result = json.loads(capsys.readouterr().out)
assert result["execute"] is False
assert not (tmp_path / "search").exists()
assert main(["select-fit", "--method", "wisp_select5", "--train", "train.npz", "--valid", "valid.npz",
"--output", str(tmp_path / "selector.pkl"), "--seed", "42", "--n-jobs", "1"]) == 0
result = json.loads(capsys.readouterr().out)
assert result["execute"] is False
assert not (tmp_path / "selector.pkl").exists()
def test_cli_evaluate_and_predict(tmp_path, capsys):
model = RecordingModel(sequence=True)
checkpoint = tmp_path / "model.pkl"
checkpoint.write_bytes(pickle.dumps(model))
(tmp_path / "checkpoint.json").write_text(json.dumps({"input_shape": [8, 3],
"label_names": ["rest", "walk"]}))
data = tmp_path / "test.npz"
np.savez(data, X=np.ones((4, 8, 3)), y=np.zeros(4, dtype=np.int64),
subject=np.array(["a", "a", "b", "b"]), time_index=np.array([0, 1, 0, 1]))
output = tmp_path / "metrics.json"
base = ["--checkpoint", str(checkpoint), "--input", str(data), "--trust-pickle"]
assert main(["evaluate", *base, "--output", str(output)]) == 0
assert json.loads(output.read_text())["macro_f1"] == 1.0
capsys.readouterr()
predictions = tmp_path / "predictions.npz"
assert main(["predict", *base, "--output", str(predictions)]) == 0
with np.load(predictions, allow_pickle=False) as archive:
assert archive["y_pred_name"].tolist() == ["rest"] * 4
capsys.readouterr()
with pytest.raises(SystemExit):
main(["predict", *base, "--output", str(predictions)])
def test_inventory_does_not_unpickle(tmp_path, capsys):
(tmp_path / "not_a_valid_model.pkl").write_bytes(b"not a pickle")
assert main(["inventory", "--root", str(tmp_path)]) == 0
result = json.loads(capsys.readouterr().out)
assert result["files_present"] == 1
assert "verified usable" in result["note"]
def test_family_constructor_preserves_names_and_forces_positive_jobs():
model = build_family("wisp_cis", seed=7, n_jobs=1, direct=True,
overrides={"n_kernels": 8, "n_intervals": 4})
assert model.seed == 7
assert model.use_hmm is False
assert model.n_kernels == 8
with pytest.raises(ValueError, match="fixed families only"):
build_family("wisp_random", seed=7)
def test_synthetic_fixed_family_round_trip(tmp_path, capsys):
random = np.random.default_rng(9)
data = tmp_path / "tiny.npz"
X = random.normal(size=(24, 24, 3))
y = np.tile([0, 2], 12).astype(np.int64)
np.savez(data, X=X, y=y, label_names=np.array(["rest", "absent", "walk"]))
output = tmp_path / "tiny.pkl"
assert main(["family-fit", "--method", "wisp_rc", "--input", str(data), "--output", str(output),
"--seed", "3", "--n-jobs", "1", "--direct", "--overrides", '{"n_kernels": 8}',
"--execute"]) == 0
capsys.readouterr()
metadata = json.loads((tmp_path / "tiny.metadata.json").read_text())
assert metadata["label_names"] == ["rest", "absent", "walk"]
fitted = load_checkpoint(output, trust_pickle=True, expected_sha256=metadata["sha256"])
assert fitted.classes_.tolist() == [0, 1, 2]
pred = predict_checkpoint(fitted, X, metadata=metadata)
assert pred.shape == (24,)
assert set(pred.tolist()) <= {0, 2}
def test_paper_worst_class_includes_unobserved_global_class():
result = score(np.array([0, 2, 0, 2]), np.array([0, 2, 0, 2]), 3)
assert result["macro_f1"] == 1.0
assert result["worst_class_f1"] == 0.0
assert len(result["confusion_matrix"]) == 3
def test_selector_tie_break_exact_precedence():
base = {"macro_f1": 0.7, "worst_class_f1": 0.4, "model_size_mb": 2.0}
def pair(**changes):
return [{**base, "method_id": "wisp_ss"}, {**base, "method_id": "wisp_rc", **changes}]
assert choose_validation_winner(pair()) == "wisp_ss"
assert choose_validation_winner(pair(model_size_mb=1.0)) == "wisp_rc"
assert choose_validation_winner(pair(worst_class_f1=0.5, model_size_mb=100.0)) == "wisp_rc"
assert choose_validation_winner(pair(macro_f1=0.8, worst_class_f1=0.0, model_size_mb=100.0)) == "wisp_rc"
with pytest.raises(ValueError, match="finite"):
choose_validation_winner(pair(macro_f1=float("nan")))
def test_selector_pickle_size_matches_preserved_default_protocol():
from wisp.cpu.algorithms import HarSculptForestHMM, estimate_pickle_size_mb
from wisp_release.selection import _model_size_mb
model = HarSculptForestHMM(seed=42, n_jobs=1)
assert _model_size_mb(model) == estimate_pickle_size_mb(model)
assert _model_size_mb(model) == len(pickle.dumps(model)) / (1024 * 1024)
@pytest.mark.parametrize("method,count", [("wisp_select5", 5), ("wisp_select7", 7)])
def test_selector_fits_train_candidates_then_refits_once_without_test(method, count, monkeypatch):
from wisp.core.data import HARData
import wisp_release.selection as selection
y = np.array([0, 1, 2, 2])
X = np.ones((4, 8, 1))
X[:, 0, 0] = y
train = HARData(X=X.copy(), y=y, subject=np.array(["train"] * 4), time_index=np.arange(4),
label_names=["a", "b", "c"])
valid = HARData(X=X.copy(), y=y, subject=np.array(["valid"] * 4), time_index=np.arange(4),
label_names=["a", "b", "c"])
SelectorFixture.fit_records = []
monkeypatch.setattr(selection, "_model_size_mb", lambda model: 1.0)
fitted, report = fit_selector(method, train, valid, seed=42, n_jobs=1,
builder=lambda family, **kwargs: SelectorFixture(family))
assert report["selected_method_id"] == "wisp_rc"
assert report["selection_uses_test"] is False
assert len(report["validation_candidates"]) == count
assert len(SelectorFixture.fit_records) == count + 1
assert all(size == 4 for _, size, _ in SelectorFixture.fit_records[:-1])
assert SelectorFixture.fit_records[-1] == ("wisp_rc", 8, 3)
assert fitted.family == "wisp_rc"
if count == 5:
assert {row["method_id"] for row in report["validation_candidates"]} == {
"wisp_ss", "wisp_rc", "wisp_so", "wisp_gis", "wisp_cis"}
leaking = HARData(X=X, y=y, subject=np.array(["train"] * 4), time_index=np.arange(4),
label_names=["a", "b", "c"])
with pytest.raises(ValueError, match="leakage"):
fit_selector(method, train, leaking, seed=42, builder=lambda family, **kwargs: SelectorFixture(family))
|