"""Read one extracted NumPy RF episode; returned RSS values are float32 dBm."""
import argparse
import json
from pathlib import Path

import numpy as np


def load_rss(directory, frames=slice(None)):
    directory = Path(directory)
    metadata = json.loads((directory / "metadata.json").read_text())
    stored = np.load(directory / metadata["training_file"], mmap_mode="r", allow_pickle=False)
    selected = stored[frames]
    quantization_file = metadata.get("quantization_file")
    if quantization_file:
        encoding = json.loads((directory / quantization_file).read_text())
        if encoding["schema"] != "uint16_centi_dbm_v1" or stored.dtype != np.dtype("uint16"):
            raise ValueError("Unsupported RSS encoding")
        values = (selected.astype(np.int32) - encoding["integer_offset"]).astype(np.float32)
        values *= np.float32(encoding["step_db"])
        values[selected == encoding["invalid_code"]] = np.nan
    else:
        if stored.dtype != np.dtype("float32"):
            raise ValueError("Expected direct float32 RSS")
        values = selected
    return np.ascontiguousarray(values)


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument("directory", type=Path)
    parser.add_argument("--frame", type=int, default=0)
    args = parser.parse_args()
    rss = load_rss(args.directory, args.frame)
    valid = rss[np.isfinite(rss)]
    print(json.dumps(dict(unit="dBm", shape=list(rss.shape), dtype=str(rss.dtype),
                          valid_voxels=int(valid.size),
                          min_dbm=float(valid.min()) if valid.size else None,
                          max_dbm=float(valid.max()) if valid.size else None)))