Retrieval Weights

Fine-tune the balance between retrieval methods for optimal results.

Default Weights

python

{
  "dense": 0.5,   # Semantic similarity
  "sparse": 0.3,  # Term expansion
  "bm25": 0.2     # Exact matching
}

Setting Weights

python

results = client.search(
    query="technical specification",
    options={
        "weights": {
            "dense": 0.4,
            "sparse": 0.3,
            "bm25": 0.3
        }
    }
)

Recommended Configurations

General Purpose

json

{ "dense": 0.5, "sparse": 0.3, "bm25": 0.2 }

Technical Documentation

json

{ "dense": 0.3, "sparse": 0.3, "bm25": 0.4 }

Conversational Queries

json

{ "dense": 0.6, "sparse": 0.3, "bm25": 0.1 }

Exact Match Priority

json

{ "dense": 0.2, "sparse": 0.2, "bm25": 0.6 }

Organization-Level Defaults

Set default weights for your organization:

python

client.settings.update({
    "default_weights": {
        "dense": 0.5,
        "sparse": 0.3,
        "bm25": 0.2
    }
})

A/B Testing Weights

Compare different configurations:

python

from lakehouse42 import ABTest

test = ABTest(
    name="weight_optimization",
    variants=[
        {"dense": 0.5, "sparse": 0.3, "bm25": 0.2},
        {"dense": 0.4, "sparse": 0.4, "bm25": 0.2},
    ]
)

results = client.search(query="...", ab_test=test)