[Question]: How to improve the accuracy on the large Milvus index?

[nithin-nk]

Question Validation

• I have searched both the documentation and discord for an answer.

Question

We are using Milvus for the vector store. We were getting very good accuracy when the index was small(1000+ records). Right now we have a collection with 200K+ records and accuracy is not good. How to improve the accuracy on the large Milvus index?

[dosubot[bot]]

🤖

Hello,

To improve the accuracy of a large Milvus index with over 200K records, you can adjust the parameters in the default_search_params dictionary in the MilvusReader class.

The nprobe parameter in the IVF_FLAT, IVF_SQ8, IVF_PQ, and IVF_HNSW index types, and the ef parameter in the HNSW, RHNSW_FLAT, RHNSW_SQ, RHNSW_PQ, and IVF_HNSW index types, and the search_k parameter in the ANNOY index type can be increased to improve the accuracy of the search results.

However, increasing these parameters will also increase the search cost (time and resources), so you need to find a balance between accuracy and efficiency based on your specific requirements.

Here is an example of how you can adjust these parameters:

self.default_search_params = {
    "IVF_FLAT": {"metric_type": "IP", "params": {"nprobe": 20}},
    "IVF_SQ8": {"metric_type": "IP", "params": {"nprobe": 20}},
    "IVF_PQ": {"metric_type": "IP", "params": {"nprobe": 20}},
    "HNSW": {"metric_type": "IP", "params": {"ef": 20}},
    "RHNSW_FLAT": {"metric_type": "IP", "params": {"ef": 20}},
    "RHNSW_SQ": {"metric_type": "IP", "params": {"ef": 20}},
    "RHNSW_PQ": {"metric_type": "IP", "params": {"ef": 20}},
    "IVF_HNSW": {"metric_type": "IP", "params": {"nprobe": 20, "ef": 20}},
    "ANNOY": {"metric_type": "IP", "params": {"search_k": 20}},
    "AUTOINDEX": {"metric_type": "IP", "params": {}},
}

In this example, the nprobe, ef, and search_k parameters have been increased from 10 to 20. You can find this in the llama_index/readers/milvus.py file.

Please note that the current default configuration for the default_search_params in the MilvusReader class in the LlamaIndex codebase is as follows:

self.default_search_params = {
    "IVF_FLAT": {"metric_type": "IP", "params": {"nprobe": 10}},
    "IVF_SQ8": {"metric_type": "IP", "params": {"nprobe": 10}},
    "IVF_PQ": {"metric_type": "IP", "params": {"nprobe": 10}},
    "HNSW": {"metric_type": "IP", "params": {"ef": 10}},
    "RHNSW_FLAT": {"metric_type": "IP", "params": {"ef": 10}},
    "RHNSW_SQ": {"metric_type": "IP", "params": {"ef": 10}},
    "RHNSW_PQ": {"metric_type": "IP", "params": {"ef": 10}},
    "IVF_HNSW": {"metric_type": "IP", "params": {"nprobe": 10, "ef": 10}},
    "ANNOY": {"metric_type": "IP", "params": {"search_k": 10}},
    "AUTOINDEX": {"metric_type": "IP", "params": {}},
}

I hope this helps! If you have any other questions, feel free to ask.

Best,
Dosu

Sources

• llama_index/readers/milvus.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.