Azure Machine Learning Model Monitoring SDK

The azureml-ai-monitoring package provides an SDK to enable Model Data Collector (MDC) for custom logging allows customers to collect data at arbitrary points in their data pre-processing pipeline. Customers can leverage SDK in score.py to log data to desired sink before, during, and after any data transformations.

Start by importing the azureml-ai-monitoring package in score.py

import pandas as pd
import json
from azureml.ai.monitoring import Collector

def init():
  global inputs_collector, outputs_collector

  # instantiate collectors with appropriate names, make sure align with deployment spec
  inputs_collector = Collector(name='model_inputs')                    
  outputs_collector = Collector(name='model_outputs')

def run(data): 
  # json data: { "data" : {  "col1": [1,2,3], "col2": [2,3,4] } }
  pdf_data = preprocess(json.loads(data))
  
  # tabular data: {  "col1": [1,2,3], "col2": [2,3,4] }
  input_df = pd.DataFrame(pdf_data)

  # collect inputs data, store correlation_context
  context = inputs_collector.collect(input_df)

  # perform scoring with pandas Dataframe, return value is also pandas Dataframe
  output_df = predict(input_df) 

  # collect outputs data, pass in correlation_context so inputs and outputs data can be correlated later
  outputs_collector.collect(output_df, context)
  
  return output_df.to_dict()
  
def preprocess(json_data):
  # preprocess the payload to ensure it can be converted to pandas DataFrame
  return json_data["data"]

def predict(input_df):
  # process input and return with outputs
  ...
  
  return output_df

Create environment with base image mcr.microsoft.com/azureml/openmpi4.1.0-ubuntu20.04 and conda dependencies, then build the environment.

channels:
  - conda-forge
dependencies:
  - python=3.8
  - numpy=1.23.5
  - pandas=1.5.2
  - pip=22.3.1
  - pip:
      - azureml-defaults==1.38.0
      - requests==2.28.1
      - azureml-ai-monitoring~=0.1.0b1
name: model-env

Create deployment with custom logging enabled (model_inputs and model_outputs are enabled) and the environment you just built, please update the yaml according to your scenario.

#source ../configs/model-data-collector/data-storage-basic-OnlineDeployment.YAML
$schema: http://azureml/sdk-2-0/OnlineDeployment.json

endpoint_name: my_endpoint #unchanged
name: blue #unchanged
model: azureml:my-model-m1:1 #azureml:models/<name>:<version> #unchanged
environment: azureml:custom-logging-env:1 #unchanged
data_collector:
  collections:
    model_inputs:
      enabled: true
    model_outputs:
      enabled: true

By default, we'll raise the exception when there is unexpected behavior (like custom logging is not enabled, collection is not enabled, not supported data type), if you want a configurable on_error, you can do it with

collector = Collector(name="inputs", on_error=lambda e: logging.info("ex:{}".format(e)))

Change Log

v0.1.0b1 (2023.4.25)

New Features

• Support model data collection for pandas Dataframe.