Implementation:SeldonIO Seldon core Seldon Pipeline Infer Monitoring

Property	Value
Implementation Name	Seldon Pipeline Infer Monitoring
Type	Wrapper Doc
Overview	Concrete tools for sending inference requests and monitoring production traffic in Seldon Core 2 pipelines
Domains	MLOps, Monitoring
Implements Principle	SeldonIO_Seldon_core_Production_Traffic_Monitoring
Source	samples/examples/income_classifier/infer.md:L28-41, docs-gb/cli/seldon_pipeline_infer.md:L1-35
External Dependencies	seldon CLI, requests, V2 protocol, Prometheus
Knowledge Sources	Repo (https://github.com/SeldonIO/seldon-core), Doc (https://docs.seldon.io/projects/seldon-core/en/v2/)
Last Updated	2026-02-13 00:00 GMT

Code Reference

CLI Inference

seldon pipeline infer income-production \
  '{"inputs": [{"name": "predict", "datatype": "FP32", "shape": [1, 12], "data": [[39,7,1,1,1,1,4,1,2174,0,40,9]]}]}'

Python Requests Inference

import requests

payload = {
    "inputs": [
        {
            "name": "predict",
            "datatype": "FP32",
            "shape": [1, 12],
            "data": [[39, 7, 1, 1, 1, 1, 4, 1, 2174, 0, 40, 9]]
        }
    ]
}

response = requests.post(
    "http://0.0.0.0:9000/v2/models/income-production/infer",
    headers={"seldon-model": "income-production.pipeline"},
    json=payload
)

print(response.json())

Expected Response Structure

{
  "model_name": "income-production",
  "outputs": [
    {
      "name": "predict",
      "datatype": "INT64",
      "shape": [1, 1],
      "data": [0]
    },
    {
      "name": "is_outlier",
      "datatype": "INT64",
      "shape": [1, 1],
      "data": [0]
    }
  ]
}

Key Parameters

Parameter	Description	Example Value
pipelineName	Name of the monitoring pipeline	income-production
inputs[].name	Input tensor name	"predict"
inputs[].datatype	Data type for features	"FP32"
inputs[].shape	Tensor shape [batch, features]	[1, 12]
inputs[].data	Feature values (12 Adult Census features)	39,7,1,1,1,1,4,1,2174,0,40,9
seldon-model header	Pipeline routing header (Python requests)	"income-production.pipeline"
URL endpoint	V2 inference endpoint	http://0.0.0.0:9000/v2/models/{pipeline}/infer

Feature Mapping (12 Adult Census Features)

Index	Feature	Example Value	Type
0	Age	39	Continuous
1	Workclass	7	Categorical
2	Education	1	Categorical
3	Marital Status	1	Categorical
4	Occupation	1	Categorical
5	Relationship	1	Categorical
6	Race	4	Categorical
7	Sex	1	Categorical
8	Capital Gain	2174	Continuous
9	Capital Loss	0	Continuous
10	Hours per Week	40	Continuous
11	Country	9	Categorical

I/O Contract

Inputs

Input	Type	Description
V2 JSON payload	JSON	Inference request with FP32 tabular features (12 dimensions)
Pipeline name	String	Target monitoring pipeline (income-production)
Seldon-model header	HTTP header	Pipeline routing (Python requests only)

Outputs

Output	Type	Description
Predictions	INT64 tensor	Classifier output (0 = <=50K, 1 = >50K)
is_outlier	INT64 tensor	Outlier flag (0 = normal, 1 = outlier)
Drift results	Async batch	Drift detection results reported per batch of 20 requests
Prometheus metrics	Time-series	Available at /metrics endpoint for monitoring dashboards

Usage Examples

Sending Normal Traffic

# Single inference request via CLI
seldon pipeline infer income-production \
  '{"inputs": [{"name": "predict", "datatype": "FP32", "shape": [1, 12], "data": [[39,7,1,1,1,1,4,1,2174,0,40,9]]}]}'

Sending Batch Traffic for Drift Detection

import requests

# Send 20+ requests to trigger drift detection batch
for i in range(25):
    payload = {
        "inputs": [
            {
                "name": "predict",
                "datatype": "FP32",
                "shape": [1, 12],
                "data": [[39, 7, 1, 1, 1, 1, 4, 1, 2174, 0, 40, 9]]
            }
        ]
    }
    response = requests.post(
        "http://0.0.0.0:9000/v2/models/income-production/infer",
        headers={"seldon-model": "income-production.pipeline"},
        json=payload
    )
    result = response.json()
    print(f"Request {i}: prediction={result['outputs'][0]['data']}, "
          f"outlier={result['outputs'][1]['data']}")

Testing with Outlier Data

# Send data that is likely to be flagged as an outlier (extreme feature values)
seldon pipeline infer income-production \
  '{"inputs": [{"name": "predict", "datatype": "FP32", "shape": [1, 12], "data": [[99,1,1,1,1,1,1,1,99999,99999,99,1]]}]}'

Related Pages

• SeldonIO_Seldon_core_Production_Traffic_Monitoring (principle) - Principles of production traffic monitoring
• SeldonIO_Seldon_core_Seldon_Pipeline_Load_And_Status (prerequisite) - Pipeline must be validated before sending traffic
• SeldonIO_Seldon_core_Seldon_Pipeline_CRD_Monitoring (pipeline definition) - The pipeline definition being invoked
• SeldonIO_Seldon_core_Alibi_Detect_Training (detector training) - How the detectors were trained
• SeldonIO_Seldon_core_Seldon_Pipeline_Version_Update (related) - Updating pipeline versions while monitoring
• Environment:SeldonIO_Seldon_core_Kafka_Messaging_Environment