import numpy as np

def dense_only_retrieve(query_emb, doc_embeddings, top_k=1000):
    sims = np.dot(doc_embeddings, query_emb)
    top_indices = np.argsort(sims)[::-1][:top_k]
    return [(int(i), float(sims[i])) for i in top_indices]

def ivf_retrieve(query_emb, cluster_manager, doc_embeddings, top_clusters=5):
    c_sims = np.dot(cluster_manager.centroids, query_emb)
    top_c = np.argsort(c_sims)[::-1][:top_clusters]
    
    candidate_docs = []
    for c in top_c:
        candidate_docs.extend(cluster_manager.cluster_to_docs[c])
        
    if not candidate_docs:
        return []
        
    candidate_embs = doc_embeddings[candidate_docs]
    sims = np.dot(candidate_embs, query_emb)
    
    results = [(candidate_docs[i], float(sims[i])) for i in range(len(candidate_docs))]
    results.sort(key=lambda x: x[1], reverse=True)
    return results

def rerank_retrieve(query_emb, sparse_results, doc_embeddings, top_sparse=1000):
    candidates = [r[0] for r in sparse_results[:top_sparse]]
    if not candidates: return []
    
    candidate_embs = doc_embeddings[candidates]
    sims = np.dot(candidate_embs, query_emb)
    
    results = [(candidates[i], float(sims[i])) for i in range(len(candidates))]
    results.sort(key=lambda x: x[1], reverse=True)
    return results