lancedb/lancedb · error · ValueError

vector_results should be a list of pa.Table or…

Error message

vector_results should be a list of pa.Table or LanceVectorQueryBuilder

What it means

After normalizing LanceVectorQueryBuilder inputs, rerank_multivector validates that each element is a pyarrow Table. Any element that is neither a query builder nor a pa.Table (e.g. a list of dicts or a pandas DataFrame) cannot be reranked, so ValueError is raised.

Solutions

1. Convert inputs to pyarrow Tables before reranking: pa.Table.from_pandas(df) or call .to_arrow() on the query builder.
2. Only pass results produced by vector search .with_row_id(True).to_arrow() (or the builders themselves).
3. Validate with all(isinstance(v, pa.Table) for v in vector_results) before the call.

Example fix

// before
reranker.rerank_multivector(query, [df1, df2])
// after
import pyarrow as pa
reranker.rerank_multivector(query, [pa.Table.from_pandas(df1), pa.Table.from_pandas(df2)])

if not all(isinstance(v, pa.Table) for v in vector_results):
    raise TypeError('each vector result must be a pyarrow Table')

def is_pyarrow_table(v):
    return isinstance(v, pa.Table)

try:
    ranked = reranker.rerank_multivector(query, vector_results)
except ValueError:
    vector_results = [v if isinstance(v, pa.Table) else pa.Table.from_pandas(v) for v in vector_results]
    ranked = reranker.rerank_multivector(query, vector_results)

When it happens

Trigger: Passing items like pandas DataFrames, dicts, or LanceDB query results wrapped in another container as vector_results elements, when the first element also isn't a LanceVectorQueryBuilder.

Common situations: Converting Arrow results to pandas for preprocessing and then passing them back to the reranker; passing raw query response objects from a different API version.

Understand the failure class

Background: Type mismatch errors: IllegalArgumentException, TypeError and type guards across 150 open-source libraries — this error's family across 150 libraries.

Related errors

• vector_results should be a list of pa.Table or…
• All elements in vector_results should be of the same type
• All elements in vector_results should be of the same type
• All elements in vector_results should be of the same type
• rerank_hybrid must return a pyarrow.Table, got

AI-assisted analysis of lancedb/lancedb@c7b051aff7 (2026-09-08). Data as JSON: /api/errors/7cdb6e9dd23bf232. Report an issue: GitHub.

Appendix: source

Thrown at python/python/lancedb/rerankers/mrr.py:140

        Reranks the results from multiple vector searches using MRR algorithm.
        Each vector search result is treated as a separate ranking system,
        and MRR calculates the mean of reciprocal ranks across all systems.
        This cannot reuse rerank_hybrid because MRR semantics require treating
        each vector result as a separate ranking system.
        """
        if not vector_results:
            raise ValueError("vector_results must not be empty")

        if not all(isinstance(v, type(vector_results[0])) for v in vector_results):
            raise ValueError(
                "All elements in vector_results should be of the same type"
            )

        # avoid circular import
        if type(vector_results[0]).__name__ == "LanceVectorQueryBuilder":
            vector_results = [result.to_arrow() for result in vector_results]
        elif not isinstance(vector_results[0], pa.Table):
            raise ValueError(
                "vector_results should be a list of pa.Table or LanceVectorQueryBuilder"
            )

        if not all("_rowid" in result.column_names for result in vector_results):
            raise ValueError(
                "'_rowid' is required for deduplication. \
                    add _rowid to search results like this: \
                    `search().with_row_id(True)`"
            )

        mrr_score_map = defaultdict(list)

        for result_table in vector_results:
            result_ids = result_table["_rowid"].to_pylist()
            for rank, result_id in enumerate(result_ids, 1):
                reciprocal_rank = 1.0 / rank
                mrr_score_map[result_id].append(reciprocal_rank)

View on GitHub (pinned to c7b051aff7)