examples/inference.py

#!/usr/bin/env python3
"""Extract L2-normalized scRep cell embeddings from an .h5ad file."""
from __future__ import annotations

import argparse
import sys
from pathlib import Path
from types import SimpleNamespace

import numpy as np

RELEASE_ROOT = Path(__file__).resolve().parents[1]
if str(RELEASE_ROOT) not in sys.path:
    sys.path.insert(0, str(RELEASE_ROOT))
from scRep_inference import encode_embeddings, load_examples, load_scRep_bundle
from scRep_pretrain.vocab import gene_vocab_to_map


def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument("--input_h5ad", required=True, help="Input AnnData (.h5ad) file.")
    parser.add_argument("--output", default="embeddings.npy", help="Output .npy embedding matrix.")
    parser.add_argument("--checkpoint", default="checkpoints/scRep_20260625_30M")
    parser.add_argument("--device", default="cuda" if __import__("torch").cuda.is_available() else "cpu")
    parser.add_argument("--batch_size", type=int, default=128)
    parser.add_argument("--max_input_genes", type=int, default=2048)
    parser.add_argument("--n_bins", type=int, default=50)
    parser.add_argument("--max_cells", type=int, default=0, help="0 means all cells.")
    parser.add_argument("--use_raw", action="store_true", help="Use adata.raw instead of adata.X.")
    return parser.parse_args()


def main() -> None:
    args = parse_args()
    checkpoint = Path(args.checkpoint)
    if not checkpoint.is_absolute():
        checkpoint = RELEASE_ROOT / checkpoint
    model, gene_vocab, *_ = load_scRep_bundle(checkpoint, SimpleNamespace(model_dir=str(checkpoint), asset_dir="", device=args.device))
    gene_name_to_id = gene_vocab_to_map(gene_vocab)
    examples, _ = load_examples(args.input_h5ad, gene_name_to_id, max_total_cells=args.max_cells, use_raw=args.use_raw)
    if not examples:
        raise ValueError("No input cells had genes present in the checkpoint vocabulary.")
    embeddings = encode_embeddings(model, examples, args)
    output = Path(args.output)
    output.parent.mkdir(parents=True, exist_ok=True)
    np.save(output, embeddings)
    print(f"Saved {embeddings.shape[0]} x {embeddings.shape[1]} embeddings to {output}")


if __name__ == "__main__":
    main()