Data Scientists do more than build fancy AI and machine learning models. They often times need to get involved with the data acquisition process. It is common for data to be pulled from other databases or even an API. Plus, the models need to be deployed. These tasks fall to the data scientist to solve (unless there is a data engineer willing to help). Recently, I have discovered Azure Functions to be an extremely useful tool for solving these types of tasks.
What are Azure Functions?
Simply stated, Azure Functions are pieces of code that run. More formally stated,
Azure Functions are serverless computing which allows code to run on-demand without the need to manage servers or hardware infrastructure.
This is exactly what a data scientist needs to solve the tasks mentioned above. I, for one, do not enjoy managing servers (hardware or virtual). I have done it before, but I find it time-consuming and tedious. It is just not my thing. Thus, I happily welcome the serverless capabilities of Azure Functions. I just focus on the code and get the task completed.
Because the code does not always need to be running, Azure Functions invoke the code based upon specified triggers. Once the trigger is activated, the code will begin to run. The following list provides some examples of the triggers available.
Triggers for Azure Functions:
- Timer – Set a timer to run the Azure Function as often as you like. Timing is specified with a cron expression.
- HTTP Rest call – Have some other code fire off an HTTP request to start the Azure Function.
- Blob storage – Run the Azure Function whenever a new file is added to a Blob storage account.
- Event Hubs – Event Hubs are often used for collecting real-time data, and this integration offers Azure Functions the ability to run when a real-time event occurs.
- Others – Cosmos DB, Service Bus, IoT Hub, GitHub are other events which can trigger an Azure Function.
What can Azure Functions Do?
Once you begin to understand the concept, you can quickly see some of the possibilities. Without having to configure servers or virtual machines, the following tasks become much simpler:
- Reading and writing data from a database
- Processing images
- Interacting with an HTTP endpoint
- Automating decisions in real-time
- Computing descriptive statistics
- Creating your own endpoint for other data scientists to call
- Automatically analyzing code after commits
Programming Languages for Azure Functions
As of August 2018, full support is provided for C#, Javascript, and F#. Experimental support is provided for Batch, PowerShell, Python, and TypeScript. Python can be used to create an HTTP endpoint. This would allow someone to quickly create an endpoint for running machine learning models via scikit-learn or another python module. Unfortunately, R is not yet available, but Microsoft has a lot invested in R, so I am expecting this eventually.
Simplify Tasks for Data Science
Next time you have a data science task which requires a little coding, consider using an Azure Function to run the code. It will most likely save you some deployment and configuration time. Then you can quickly get back to optimizing those fancy AI models.
See the video below for a quick demonstration of how to create an Azure function via a web browser (no IDE needed).
