Data Science

Defining a Machine Learning Workflow

Deploy a model
- We can use AzureML or Databricks to achieve this
Build a pipeline
- Pipelines are used to automate model deploying in the future
- We will typically want to automatically tune the model
- We can build a pipeline to avoid repeating these steps:
  1. Preparing more data
  2. Validating the data
  3. Deploying the model manually
- We would use the AzureML or Azure Devops to achieve this

Example of Model Deployment in Azure

Create a workspace in the Azure Machine Learning service
Create a model using regular Python code in any IDE
- This could include creating a model using sklearn, tensorflow, etc.
- We would want to save this model using pickle
Initialize the workspace using the Python SDK in an IDE

from azureml.core import Workspace
ws = Workspace.create(
	name='myworkspace',
	subscription_id='<azure-subscription-id>',
	resource_group='myresourcegroup',
	create_resource_group=True,
	location='eastus2'
	)

Register input data and output data using the Python SDK in an IDE
- We want to record the data used to create this model in our workspace

from azureml.core import Dataset
datastore = ws.get_default_datastore()
datastore.upload_files(
	files=['./features.csv', './labels.csv'],
	target_path='sklearn_regression/',
	overwrite=True
	)
input_dataset = Dataset.Tabular.from_delimited_files(path=[(datastore, 'sklearn_regression/features.csv')])
output_dataset = Dataset.Tabular.from_delimited_files(path=[(datastore, 'sklearn_regression/labels.csv')])

from azureml.core import Model
from azureml.core.resource_configuration import ResourceConfiguration
model = Model.register(
	workspace=ws,
	model_name='my-sklearn-model',
	model_path='./sklearn_regression_model.pkl',
	model_framework=Model.Framework.SCIKITLEARN,
	model_framework_version='0.19.1',
	sample_input_dataset=input_dataset,
	sample_output_dataset=output_dataset,
	resource_configuration=ResourceConfiguration(cpu=1, memory_in_gb=0.5),
	description='Ridge regression model to predict diabetes progression.',
	tags={'area': 'diabetes', 'type': 'regression'}
	)

Deploy our model using the Python SDK in an IDE
- Our model is deployed as a containerized web app (or web service)
- In other words, Azure deploys our model by building an image (from the model) and running containers based on that image
- For a deployed web service, we can choose from the following container instances:
  - Azure Container Instance (ACI)
  - FPGA
  - Docker
- If we wanted to run our web service locally (rather than running it on Azure ML service), then we would want to deploy our model to a local Docker container instead
- If we wanted to run a more production-ready web service on Azure ML, then we want to deploy our model to the Azure Kubernetes service instead

from azureml.core import Webservice
service = Model.deploy(ws, 'my-sklearn-service', [model])

Run the deployed web service using the Python SDK in an IDE

import json
input_payload = json.dumps({
	'data': [[1, 2, 3, 4, 5, 4, 3, 2, 9, 187]],
	'method': 'predict'
	})
output = service.run(input_payload)

Defining Runs and Experiments in Azure

The Run class is in AzureML
- It refers to Python code for a specific task
- For example, training a model or tuning hyperparameters
A run logs metrics and uploads the result to the Azure platform
- It's a more natural way to track jobs in a workspace
The Experiment class is a term that refers to a composition of a series of runs
- For example, we could have one run for a logistic regression model
- Another run could be for a kNN model
- Together, these runs could make up an experiment
We create runs and experiments to easily monitor and review a model throughout its training, testing, or tuning process
We can observe details about runs or experiments in the AzureML
- By going to the workspace associated with the runs or experiments

from azureml.core import Experiment
#Create an experiment
experiment = Experiment(workspace = ws, name = "my-first-experiment")
#Create a run
run = experiment.start_logging()
run.log("trial",1)
run.complete()

Other Model Deployment Options

We can deploy an image already created from a model
In this case, we would use the deploy_from_image method from the Webservice class
We could also use the deploy method from the Webservice class
We can deploy a model already registered in the workspace
In this case, we could use the deploy_from_model method from the Webservice class
We could also use the deploy method from the Model class
This method will create an image

References

Azure Workspaces

Azure AutoML

Azure Model Deployment

Defining a Machine Learning Workflow

Example of Model Deployment in Azure

Defining Runs and Experiments in Azure

Other Model Deployment Options

References