向量数据库2026选型：Milvus vs Pinecone vs Weaviate vs Qdrant

# === Milvus 混合检索 ===
from pymilvus import Collection

collection.search(
    data=[query_vector],
    anns_field="embedding",
    param={"metric_type": "COSINE", "params": {"nprobe": 16}},
    expr='source == "tech_blog" and date > "2026-01-01"',  # 标量过滤
    limit=10,
)

# === Pinecone 混合检索 ===
from pinecone import Pinecone

index.query(
    vector=query_vector,
    filter={"source": {"$eq": "tech_blog"}, "date": {"$gt": "2026-01-01"}},
    top_k=10,
    include_metadata=True,
)

# === Weaviate 混合检索（原生支持BM25+Vector） ===
import weaviate

result = client.query.get("Document", ["content", "title"]) \
    .with_hybrid(
        query="GraphRAG实践指南",
        alpha=0.5,  # 0=纯BM25, 1=纯向量
    ) \
    .with_limit(10) \
    .do()

# === Qdrant 混合检索 ===
from qdrant_client import QdrantClient

client.search(
    collection_name="documents",
    query_vector=query_vector,
    query_filter={
        "must": [
            {"key": "source", "match": {"value": "tech_blog"}},
            {"key": "date", "range": {"gt": "2026-01-01"}},
        ]
    },
    limit=10,
)

# Pinecone：5分钟上手
from pinecone import Pinecone, ServerlessSpec

pc = Pinecone(api_key="your-key")
index = pc.create_index(
    name="rag-docs",
    dimension=1536,
    spec=ServerlessSpec(cloud="aws", region="us-east-1"),
)

index.upsert(vectors=[{"id": "1", "values": [0.1, ...]}])
results = index.query(vector=[0.1, ...], top_k=5)

from pymilvus import connections, Collection, FieldSchema, CollectionSchema, DataType

# 连接Milvus集群
connections.connect(host="milvus-cluster.internal", port="19530")

# 创建Collection
fields = [
    FieldSchema(name="id", dtype=DataType.INT64, is_primary=True),
    FieldSchema(name="embedding", dtype=DataType.FLOAT_VECTOR, dim=768),
    FieldSchema(name="metadata", dtype=DataType.JSON),
]
schema = CollectionSchema(fields, "large_scale_rag")
collection = Collection("documents", schema)

# 创建索引
collection.create_index(
    "embedding",
    {
        "index_type": "HNSW",
        "metric_type": "COSINE",
        "params": {"M": 48, "efConstruction": 512},
    },
)

# 分区策略：按时间分区
collection.create_partition("2026_06")
collection.create_partition("2026_07")

import weaviate

client = weaviate.connect_to_local()

# 创建Schema
client.collections.create(
    name="Document",
    vectorizer_config=Configure.Vectorizer.text2vec_openai(),
    properties=[
        Property(name="content", data_type=DataType.TEXT),
        Property(name="title", data_type=DataType.TEXT),
        Property(name="category", data_type=DataType.TEXT),
    ],
)

# 混合检索：向量+BM25
collection = client.collections.get("Document")
results = collection.query.hybrid(
    query="GraphRAG实践指南",
    query_properties=["content", "title"],
    alpha=0.5,  # 0=BM25, 1=向量
    limit=10,
)

from qdrant_client import QdrantClient
from qdrant_client.models import Distance, VectorParams, PointStruct

client = QdrantClient(host="localhost", port=6333)

# 创建Collection（启用量化降低内存）
client.create_collection(
    collection_name="documents",
    vectors_config=VectorParams(size=768, distance=Distance.COSINE),
    quantization_config=ScalarQuantization(
        scalar=ScalarQuantizationConfig(
            type=ScalarType.INT8,
            quantile=0.99,
            always_ram=True,
        ),
    ),
)

# 搜索（启用量化加速）
results = client.search(
    collection_name="documents",
    query_vector=query_embedding,
    search_params=SearchParams(
        quantization=QuantizationSearchParams(
            ignore=False,
            rescore=True,  # 先量化粗排，再精确重排
        ),
        hnsw_ef=128,
    ),
    limit=10,
)