code/spec_rag/retrieval.py

from __future__ import annotations

from typing import Dict, List, Tuple

import numpy as np

from .faiss_index import index_search


def search_index(index, query_embeddings: np.ndarray, k: int) -> Tuple[np.ndarray, np.ndarray]:
    scores, indices = index_search(index, query_embeddings, k)
    return scores, indices


def recall_at_k(
    retrieved_indices: np.ndarray,
    ground_truth_indices: List[int],
    k: int,
) -> float:
    if len(ground_truth_indices) == 0:
        return 0.0
    hits = 0
    total = 0
    for row, gt in zip(retrieved_indices, ground_truth_indices):
        if gt < 0:
            continue
        total += 1
        if gt in row[:k]:
            hits += 1
    if total == 0:
        return 0.0
    return hits / float(total)


def build_smiles_to_index(smiles: List[str]) -> Dict[str, int]:
    lookup: Dict[str, int] = {}
    for i, smi in enumerate(smiles):
        if smi not in lookup:
            lookup[smi] = i
    return lookup