Conditional conformal selection¶

import numpy as np
from sklearn.ensemble import IsolationForest

from nonconform import ConformalDetector, Split
from nonconform.metrics import false_discovery_rate, statistical_power
from nonconform.scoring import ConditionalEmpirical

rng = np.random.default_rng(42)
x_reference = rng.normal(size=(700, 4))
x_test = np.vstack(
    [rng.normal(size=(15, 4)), rng.normal(loc=4.5, size=(5, 4))]
)
y_test = np.r_[np.zeros(15, dtype=int), np.ones(5, dtype=int)]

for method in ["mc", "simes", "dkwm", "asymptotic"]:
    estimation_kwargs = {
        "method": method,
        "delta": 0.1,
        "tie_break": "classical",
    }
    if method == "mc":
        estimation_kwargs["mc_num_simulations"] = 500

    detector = ConformalDetector(
        detector=IsolationForest(n_estimators=50, random_state=42),
        strategy=Split(n_calib=0.3),
        estimation=ConditionalEmpirical(**estimation_kwargs),
        seed=42,
    ).fit(x_reference)

    selected = np.asarray(detector.select(x_test, alpha=0.1))
    result = detector.last_result
    assert result is not None
    assert result.p_values is not None

    print(
        method,
        {
            "minimum_p": float(result.p_values.min()),
            "discoveries": int(selected.sum()),
            "realized_fdp": float(false_discovery_rate(y_test, selected)),
            "power": float(statistical_power(y_test, selected)),
        },
    )