task_repo/solution.py

#!/usr/bin/env python3
"""Deliberately modest deterministic baseline; agents should improve it."""
from __future__ import annotations

import argparse
from pathlib import Path
import sys

import numpy as np
from sklearn.linear_model import LogisticRegression
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler


def _features(signal: np.ndarray, rr: np.ndarray) -> np.ndarray:
    # Coarse morphology intentionally leaves substantial room for better signal models.
    n, channels, width = signal.shape
    bins = 18
    trimmed = signal[:, :, : (width // bins) * bins]
    chunks = trimmed.reshape(n, channels, bins, -1)
    means = chunks.mean(axis=3).reshape(n, -1)
    stds = chunks.std(axis=3).reshape(n, -1)
    extrema = np.concatenate(
        [signal.min(axis=2), signal.max(axis=2), np.ptp(signal, axis=2)], axis=1
    )
    return np.concatenate([rr, means, stds, extrema], axis=1).astype(np.float32)


def run(train_path: Path, test_path: Path, output_path: Path) -> None:
    with np.load(train_path, allow_pickle=False) as train:
        x_train = _features(train["signal"], train["rr"])
        y_train = train["y"].astype(np.int64)
    with np.load(test_path, allow_pickle=False) as test:
        x_test = _features(test["signal"], test["rr"])

    keep = y_train < 4  # Q is too sparse for the baseline.
    model = make_pipeline(
        StandardScaler(),
        LogisticRegression(
            C=0.7,
            class_weight="balanced",
            max_iter=250,
            random_state=20260920,
            solver="lbfgs",
        ),
    )
    model.fit(x_train[keep], y_train[keep])
    pred = model.predict(x_test).astype(np.int16)
    output_path.parent.mkdir(parents=True, exist_ok=True)
    np.save(output_path, pred, allow_pickle=False)


def self_test() -> None:
    rng = np.random.default_rng(7)
    sig = rng.normal(size=(12, 2, 252)).astype(np.float32)
    rr = rng.normal(size=(12, 5)).astype(np.float32)
    got = _features(sig, rr)
    assert got.shape == (12, 83), got.shape
    assert np.isfinite(got).all()
    print("self-test passed")


def main() -> int:
    parser = argparse.ArgumentParser()
    parser.add_argument("--train", type=Path)
    parser.add_argument("--test", type=Path)
    parser.add_argument("--output", type=Path)
    parser.add_argument("--self-test", action="store_true")
    args = parser.parse_args()
    if args.self_test:
        self_test()
        return 0
    if not (args.train and args.test and args.output):
        parser.error("--train, --test and --output are required")
    run(args.train, args.test, args.output)
    return 0


if __name__ == "__main__":
    sys.exit(main())