Download tests/test_model_c.py from efe-T/nasa-same-model-c: direct link, hf CLI and curl.
- Browser
- Download file 5.54 kB
-
https://huggingface.co/efe-T/nasa-same-model-c/resolve/main/tests/test_model_c.py
- Command line
-
hf download hf://efe-T/nasa-same-model-c/tests/test_model_c.py
-
curl -L -o test_model_c.py https://huggingface.co/efe-T/nasa-same-model-c/resolve/main/tests/test_model_c.py
5.54 kB
| import copy | |
| import itertools | |
| import json | |
| from pathlib import Path | |
| import joblib | |
| import numpy as np | |
| import pandas as pd | |
| import pytest | |
| from model_c.data import acquire, measurement, number, preprocess, WORKBOOK_SHA, sha | |
| from model_c.inference import FEATURES, NUMERIC, UNITS, load_bundle, predict_model_c | |
| def request_data(): | |
| return json.loads(Path("reports/sample_inference.json").read_text())["request"] | |
| def test_missing_not_zero(value): | |
| assert number(value)[0] is None | |
| def test_detection_limit_and_saturation_distinct(): | |
| assert measurement(0)[0] == 0 | |
| assert measurement(.01, below_limit=True) == (None, "below_detection_limit") | |
| assert measurement(9, saturated=True) == (None, "saturated") | |
| assert measurement(None)[1] not in ["saturated", "below_detection_limit"] | |
| def test_source_integrity(): | |
| assert sha(acquire()) == WORKBOOK_SHA | |
| def test_preprocessing_reproducible(tmp_path): | |
| a = preprocess(acquire(), tmp_path) | |
| b = pd.read_csv("data/processed/observations.csv") | |
| assert a.source_row.tolist() == b.source_row.tolist() | |
| np.testing.assert_allclose(a.iss_scatter_v, b.iss_scatter_v) | |
| assert set(a.campaign) == {"SAME", "SAME-R"} | |
| assert "DBP" not in set(a.material) | |
| assert a[a.source_test_id.str.contains(r"\d+\s*a", case=False, regex=True)].training_eligible.sum() == 0 | |
| def test_no_group_leakage(): | |
| metrics = json.loads(Path("reports/metrics.json").read_text()) | |
| for fold in metrics["outer_folds"]: | |
| assert not set(fold["train_groups"]) & set(fold["test_groups"]) | |
| df = pd.read_csv("data/processed/observations.csv") | |
| assert df.groupby("run_id").group_id.nunique().max() == 1 | |
| assert df.groupby("sample_id").group_id.nunique().max() == 1 | |
| def test_harmonization_units_and_label(): | |
| df = pd.read_csv("data/processed/observations.csv") | |
| np.testing.assert_array_equal(df.alarm_level_reached, (df.iss_scatter_v >= 2).astype(int)) | |
| assert set(df.label_provenance) == {"derived"} | |
| assert not set(FEATURES) & {"iss_scatter_v", "alarm_level_reached", "m0_source", "m1_mm_cm3"} | |
| d = json.loads(Path("data/processed/harmonization.json").read_text()) | |
| assert d["controls"]["N"] == "airflow_cm_s" | |
| def test_export_round_trip_and_shape(tmp_path, request_data): | |
| model, domain = load_bundle() | |
| joblib.dump(model, tmp_path / "roundtrip.joblib") | |
| loaded = joblib.load(tmp_path / "roundtrip.joblib") | |
| x = pd.DataFrame([{k: request_data[k] for k in FEATURES}]) | |
| assert model.predict(x).shape == (1,) | |
| np.testing.assert_array_equal(loaded.predict(x), model.predict(x)) | |
| assert predict_model_c(request_data)["value_kind"] == "predicted" | |
| def test_missing_fields_rejected(field, request_data): | |
| request_data.pop(field) | |
| with pytest.raises(ValueError): | |
| predict_model_c(request_data) | |
| def test_invalid_numeric(bad, request_data): | |
| request_data["airflow_cm_s"] = bad | |
| with pytest.raises(ValueError): | |
| predict_model_c(request_data) | |
| def test_unseen_material(request_data): | |
| request_data["material"] = "PMMA" | |
| with pytest.raises(ValueError, match="material"): | |
| predict_model_c(request_data) | |
| def test_wrong_units(request_data): | |
| request_data["units"] = dict(UNITS, airflow_cm_s="m/s") | |
| with pytest.raises(ValueError, match="units"): | |
| predict_model_c(request_data) | |
| def test_out_of_range(key, request_data): | |
| _, domain = load_bundle() | |
| request_data[key] = domain["ranges"][key][1] + 1 | |
| with pytest.raises(ValueError, match="outside"): | |
| predict_model_c(request_data) | |
| def test_accepted_ood_exists(): | |
| example = json.loads(Path("reports/ood_example.json").read_text()) | |
| response = predict_model_c(example["request"]) | |
| assert response["ood"] is True | |
| assert response["nearest_distance"] > response["ood_threshold"] | |
| def test_ood_threshold_different_group(): | |
| _, domain = load_bundle() | |
| df = pd.DataFrame(domain["training_conditions"]) | |
| maximum = 0 | |
| for _, row in df.iterrows(): | |
| other = df[(df.material == row.material) & (df.group_id != row.group_id)] | |
| distances = np.linalg.norm((other[NUMERIC].to_numpy() - row[NUMERIC].to_numpy(dtype=float)) / domain["scale"], axis=1) | |
| maximum = max(maximum, distances.min()) | |
| assert domain["ood_threshold"] == pytest.approx(maximum) | |
| def test_scientific_formula_agreement(): | |
| metrics = json.loads(Path("reports/metrics.json").read_text()) | |
| assert metrics["scientific_checks"]["workbook_moment_ratio_max_absolute_error_um"] < 1e-12 | |
| def test_external_source_discrepancy_preserved(): | |
| df = pd.read_csv("reports/external_checks.csv") | |
| failures = df[df.agreement_status == "unresolved_source_table_inconsistency"] | |
| assert list(zip(failures.material, failures.state)) == [("Teflon", "unaged")] | |
| assert failures.iloc[0].source_count_um == .101 | |
| assert failures.iloc[0].calculated_count_um > .11 | |
| def test_dashboard_has_all_observations(): | |
| a = pd.read_csv("data/processed/observations.csv") | |
| b = pd.read_csv("data/processed/dashboard_table.csv") | |
| assert len(a) == len(b) | |
| assert set(b.derived_label_kind) == {"derived"} | |
| assert b[b.training_eligible].oof_predicted.notna().all() | |
| assert b[~b.training_eligible].predicted_alarm_level_reached.isna().all() | |