Implementation:SeldonIO Seldon core PandasQuery Model

Knowledge Sources	SeldonIO_Seldon_core MLServer
Domains	Data_Filtering, Pipeline_Routing, ML_Serving
Last Updated	2026-02-13 14:00 GMT

Overview

Custom MLServer model that filters tabular data using pandas DataFrame query expressions for conditional pipeline routing in Seldon Core 2.

Description

The PandasQueryRuntime class extends MLModel from MLServer to implement a data filtering step. On load, it reads a query expression from the model settings parameters (settings.parameters.extra["query"]). During inference, it decodes the input as a pandas DataFrame using PandasCodec, applies the query expression via DataFrame.query(), and returns the filtered rows with a status column. If no rows match the query, a status message is returned instead.

This is used in Seldon Core 2 pipelines for conditional routing, where a model step filters data before passing it downstream.

Usage

Use this model as a pipeline step when you need to filter incoming tabular data based on a pandas query expression. Configure the query via model settings in the model YAML or model-settings.json file.

Code Reference

Source Location

• Repository: SeldonIO_Seldon_core
• File: samples/examples/pandasquery/pandasquery/model.py
• Lines: 1-47

Signature

class ModelParametersMissing(MLServerError):
    def __init__(self, model_name: str, reason: str):
        """Raised when required model parameters are missing."""

class PandasQueryRuntime(MLModel):
    async def load(self) -> bool:
        """Load query from settings.parameters.extra['query']."""

    async def predict(self, payload: InferenceRequest) -> InferenceResponse:
        """Filter input DataFrame using the configured pandas query expression."""

Import

from pandasquery.model import PandasQueryRuntime

I/O Contract

Inputs

Name	Type	Required	Description
payload	InferenceRequest	Yes	V2 inference request with tabular data decodable by PandasCodec
settings.parameters.extra["query"]	string	Yes	Pandas DataFrame query expression (e.g., "Age > 30")

Outputs

Name	Type	Description
InferenceResponse	InferenceResponse	Filtered DataFrame encoded via PandasCodec with added "status" column

Usage Examples

Model Settings Configuration

{
    "name": "pandasquery",
    "implementation": "pandasquery.model.PandasQueryRuntime",
    "parameters": {
        "extra": {
            "query": "Age > 30"
        }
    }
}

Seldon Model YAML

apiVersion: mlops.seldon.io/v1alpha1
kind: Model
metadata:
  name: pandasquery
spec:
  storageUri: "gs://seldon-models/pandasquery"
  requirements:
    - pandas
  parameters:
    - name: query
      value: "Age > 30"

Pipeline with PandasQuery

apiVersion: mlops.seldon.io/v1alpha1
kind: Pipeline
metadata:
  name: income-filter
spec:
  steps:
    - name: income
    - name: pandasquery
      inputs:
        - income
  output:
    steps:
      - pandasquery

Related Pages

• Environment:SeldonIO_Seldon_core_Python_ML_Dependencies_Environment