ssegui/mlops-demo

Demo for MLOps with Azure Machine Learning

mlops-demo

Demo for MLOps with Azure Machine Learning

Setup

To be written.

Conventions

This repo is fully based on conventions in order to make MLOps reusable and easily scaleable. The directory structure is as follows:

pipelines
    \- train-and-register.yml - Base pipeline for training and registering a model
models
    \- model1
        train.py (entry file for training)
        score.py (entry file for scoring)
        \- inference-config
            conda_inference.yml - Conda environement definition for inference/scoring
            inference-config.json - Azure Machine Learning config for inferencing
            prod-deployment-config.yml - Production deployment infrastructure definition (e.g., AKS configuration)
            qa-deployment-config.yml - QA deployment infrastructure definition (e.g., ACI configuration)
        \- train-config
            conda_train.yml - Conda environement definition for training
    \- model2
        ...same file and folder structure...

Testing

This snipped can be used to manually showcase/test the deployed model on ACI:

import requests
import json

url = '<scoring url>'

test_sample = json.dumps({
  'data': {
    "Age": [
      20
    ],
    "Sex": [
      "male"
    ],
    "Job": [
      0
    ],
    "Housing": [
      "own"
    ],
    "Saving accounts": [
      "little"
    ],
    "Checking account": [
      "little"
    ],
    "Credit amount": [
      100
    ],
    "Duration": [
      48
    ],
    "Purpose": [
      "radio/TV"
    ]
  }
})

test_sample = bytes(test_sample,encoding = 'utf8')

headers = {'Content-Type':'application/json'}
resp = requests.post(url, test_sample, headers=headers)

print("prediction:", resp.text)