Model Registry Tech Conference 2026 — จัดการ ML

Model Registry

Model Registry MLflow Model Versioning A/B Testing Model Serving Production MLOps Weights Biases Neptune Vertex AI Kubernetes Docker Seldon KServe

Tool	Type	Hosting	ราคา	เหมาะกับ
MLflow	Open Source	Self-host / Cloud	ฟรี	ทั่วไป
Weights & Biases	SaaS	Cloud	Free / $50/mo	Experiment Track
Neptune.ai	SaaS	Cloud	Free / Custom	Team Collab
Vertex AI	GCP Managed	Cloud	Pay-per-use	GCP Users
SageMaker	AWS Managed	Cloud	Pay-per-use	AWS Users

MLflow Setup

=== MLflow Model Registry ===

อ่านเพิ่ม: TTS Coqui Data Pipeline ETL — วิธีตั้งค่าและใช้งานจริงพร้อมต · อ่านเพิ่ม: DuckDB Analytics Edge Deployment — วิธีตั้งค่าและใช้งานจริงพ · อ่านเพิ่ม: LLM Inference vLLM CQRS Event Sourcing | SiamCafe Blog

pip install mlflow scikit-learn

import mlflow

import mlflow.sklearn

from sklearn.ensemble import RandomForestClassifier

from sklearn.datasets import make_classification

from sklearn.model_selection import train_test_split

from sklearn.metrics import accuracy_score, f1_score

import numpy as np

Generate sample data

X, y = make_classification(n_samples=1000, n_features=20, random_state=42)

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

MLflow Experiment

mlflow.set_tracking_uri("http://localhost:5000")

mlflow.set_experiment("conference-model-2026")

Train and Log Model

with mlflow.start_run(run_name="rf-v1") as run:

params = {"n_estimators": 100, "max_depth": 10, "random_state": 42}

mlflow.log_params(params)

model = RandomForestClassifier(**params)

model.fit(X_train, y_train)

y_pred = model.predict(X_test)

accuracy = accuracy_score(y_test, y_pred)

f1 = f1_score(y_test, y_pred)

mlflow.log_metrics({"accuracy": accuracy, "f1_score": f1})

mlflow.sklearn.log_model(model, "model",

registered_model_name="conference-classifier")

print(f"Run ID: {run.info.run_id}")

print(f"Accuracy: {accuracy:.4f} | F1: {f1:.4f}")

from mlflow.tracking import MlflowClient

client = MlflowClient()

# Transition to Staging

client.transition_model_version_stage(

name="conference-classifier",

version=1,

stage="Staging"

)

# After testing, promote to Production

client.transition_model_version_stage(

name="conference-classifier",

version=1,

stage="Production"

)

from dataclasses import dataclass

@dataclass

class ModelVersion:

name: str

version: int

stage: str

accuracy: float

f1: float

created_by: str

created_at: str

versions = [

ModelVersion("conference-classifier", 1, "Archived", 0.8850, 0.8820, "data-team", "2025-01-01"),

ModelVersion("conference-classifier", 2, "Archived", 0.9120, 0.9080, "data-team", "2025-01-10"),

ModelVersion("conference-classifier", 3, "Staging", 0.9350, 0.9310, "ml-engineer", "2025-01-15"),

ModelVersion("conference-classifier", 4, "Production", 0.9480, 0.9450, "ml-engineer", "2025-01-20"),

]

print("=== Model Registry ===")

for v in versions:

print(f" [v{v.version}] {v.name} — {v.stage}")

print(f" Accuracy: {v.accuracy:.4f} | F1: {v.f1:.4f}")

print(f" By: {v.created_by} | Date: {v.created_at}")

Model Serving

=== Model Serving & Deployment ===

MLflow Model Serve

mlflow models serve -m "models:/conference-classifier/Production" -p 5001

# Test

curl -X POST http://localhost:5001/invocations \

-H "Content-Type: application/json" \

-d '{"inputs": [[1.0, 2.0, 3.0, ...]]}'

Docker Deployment

mlflow models build-docker \

-m "models:/conference-classifier/Production" \

-n conference-model:latest

docker run -p 5001:8080 conference-model:latest

Kubernetes Deployment

apiVersion: apps/v1

kind: Deployment

metadata:

name: conference-model

spec:

replicas: 3

selector:

matchLabels:

app: conference-model

template:

metadata:

labels:

app: conference-model

spec:

containers:

name: model

image: conference-model:latest

ports:

containerPort: 8080

resources:

requests:

cpu: "500m"

memory: "1Gi"

limits:

cpu: "2"

memory: "4Gi"

readinessProbe:

httpGet:

path: /health

port: 8080

A/B Testing with Istio

apiVersion: networking.istio.io/v1beta1

kind: VirtualService

spec:

http:

route:
destination:

host: model-v3

weight: 80

destination:

host: model-v4

weight: 20

@dataclass

class ServingConfig:

method: str

latency_ms: int

throughput_rps: int

scaling: str

cost: str

configs = [

ServingConfig("MLflow Serve", 50, 100, "Manual", "Low"),

ServingConfig("Docker + K8s", 30, 500, "HPA Auto", "Medium"),

ServingConfig("Seldon Core", 25, 1000, "Auto + A/B", "Medium"),

ServingConfig("KServe", 20, 2000, "Serverless", "Pay-per-use"),

ServingConfig("BentoML", 15, 1500, "Auto", "Medium"),

]

print("\n=== Serving Options ===")

for s in configs:

print(f" [{s.method}]")

print(f" Latency: {s.latency_ms}ms | Throughput: {s.throughput_rps} RPS")

print(f" Scaling: {s.scaling} | Cost: {s.cost}")

MLOps Pipeline

# === MLOps Pipeline ===

# CI/CD for ML
# .github/workflows/ml-pipeline.yml
# name: ML Pipeline
# on:
#   push:
#     paths: ['models/**', 'data/**']
# jobs:
#   train:
#     runs-on: ubuntu-latest
#     steps:
#       - uses: actions/checkout@v4
#       - run: pip install -r requirements.txt
#       - run: python train.py
#       - run: mlflow models build-docker -m runs:/$RUN_ID/model -n model:$SHA
#   test:
#     needs: train
#     steps:
#       - run: python test_model.py --model-uri models:/model/Staging
#   deploy:
#     needs: test
#     if: github.ref == 'refs/heads/main'
#     steps:
#       - run: kubectl set image deployment/model model=model:$SHA

mlops_metrics = {
    "Models in Registry": "12",
    "Production Models": "4",
    "Daily Predictions": "2.5M",
    "Model Retrain Frequency": "Weekly",
    "A/B Tests Running": "2",
    "Avg Inference Latency": "25ms",
    "Model Accuracy (Production)": "94.8%",
    "Data Drift Alerts (30d)": "3",
    "Pipeline Runs (30d)": "45",
}

print("MLOps Dashboard:")
for k, v in mlops_metrics.items():
    print(f"  {k}: {v}")

conference_agenda = [
    "09:00 Keynote: MLOps in 2026 — State of the Art",
    "10:00 Workshop: MLflow Model Registry Hands-on",
    "11:30 Talk: A/B Testing ML Models at Scale",
    "13:00 Workshop: Kubernetes Model Serving with KServe",
    "14:30 Talk: Data Drift Detection & Auto-retraining",
    "16:00 Panel: MLOps Best Practices from Industry Leaders",
    "17:00 Lightning Talks: Community Projects",
]

print(f"\n\nTech Conference 2026 — ML Track:")
for item in conference_agenda:
    print(f"  {item}")

เคล็ดลับ

Version: Version ทุก Model อย่า Deploy โดยไม่ Track
A/B Test: ทดสอบ Model ใหม่ก่อน Full Deploy เสมอ
Monitor: ดู Data Drift Model Performance ทุกวัน
Rollback: ต้อง Rollback ได้ภายใน 5 นาที
Automate: CI/CD สำหรับ ML Pipeline อัตโนมัติ

Model Registry คืออะไร

จัดเก็บจัดการ ML Model รวมศูนย์ Version Metadata Metrics Stage Staging Production Rollback MLflow W&B Neptune Vertex AI

Model Registry Tech Conference 2026 — จัดการ ML

Model Registry

MLflow Setup

with mlflow.start_run(run_name="rf-v1") as run:

class ModelVersion:

for v in versions:

Model Serving

metadata:

spec:

selector:

matchLabels:

template:

metadata:

labels:

spec:

containers:

ports:

resources:

requests:

limits:

readinessProbe:

spec:

class ServingConfig:

for s in configs:

MLOps Pipeline

เคล็ดลับ

Model Registry คืออะไร

บทความที่เกี่ยวข้อง