"""Next Federation: load the released global model (FedProx) or rerun the federated simulation. Usage: python inference.py --selftest from inference import load, predict """ import json import os import sys HERE = os.path.dirname(os.path.abspath(__file__)) sys.path.insert(0, HERE) import numpy as np # noqa: E402 import fed_core as fc # noqa: E402 from nlrt import forward, load_model # noqa: E402 def load(path=os.path.join(HERE, "model.json")): return load_model(path) def predict(spec, pixels): """pixels: (N, 64) ints 0..16. Returns class probabilities (N, 10) from the global model.""" z = forward(spec["nets"]["global"], np.asarray(pixels, dtype=float) / 16.0) z = z - z.max(axis=1, keepdims=True) e = np.exp(z) return e / e.sum(axis=1, keepdims=True) def _selftest(): spec = load() with open(os.path.join(HERE, "selftest.json"), encoding="utf-8") as f: t = json.load(f) P = predict(spec, t["pixels"]) diff = float(np.abs(P - np.asarray(t["expected_proba"])).max()) # rerun a tiny federated simulation and compare with stored output X = np.asarray(t["sim"]["pixels"], dtype=float) / 16.0 y = np.asarray(t["sim"]["y"]) parts = fc.partition(y.tolist(), 4, 0.5, 2) gt = list(range(len(y))) r = fc.run(X, y, parts, X[gt], y[gt], "fedprox", rounds=2, seed=2, **spec["meta"]["hyper"]) sim_diff = float(np.abs(np.asarray(r["curve"]) - np.asarray(t["sim"]["expected_curve"])).max()) ok = diff < 1e-6 and sim_diff < 1e-9 print(json.dumps({"passed": ok, "max_abs_diff": diff, "sim_curve_diff": sim_diff, "n_samples": len(t["pixels"])})) return ok if __name__ == "__main__": if "--selftest" in sys.argv: sys.exit(0 if _selftest() else 1) print(__doc__)