Learn about what Azure Machine Learning is and the value it brings.
- [Instructor] Let's start with a little description of what is Azure ML? Now, even though in this course I will not be diving into the details of things like creating a model using Azure ML, I have other courses on the LinkedIn Learning that go into that topic well in depth but I think it is important to know what are we talking about here. When you look at the spectrum of AI, especially within the Microsoft ecosystem, you have, I don't what exactly call it the lower end but let's call it the easy end because these are quite powerful.
Cognitive services. Cognitive services, they're just REST API's, probably the most important selling point of cognitive services is that you don't need to be an AI developer to take advantage of cognitive services. But think of it as a SAS application, so cognitive services are powerful but they are what they are. You can't change their definition, you can't really add a new cognitive service, it's up to Microsoft to do that.
But certainly within the bounds of cognitive services, a lot of interesting applications could be built. What if you need more than that? For that is where custom AI comes in where you have full control. You can write your own algorithms, you can develop and execute where you want. But yeah, you need to be an AI expert, there's a fairly significant learning curve there and requires a lot of set up. You may need access to a powerful local machine, resources in the cloud, most importantly how to manage all of that.
You know, how you can execute a model locally and in the cloud and sort of scale as you grow, lot of concepts learned there. So there's a lot of installation and setup and learning before you dive into custom AI. Azure ML sits somewhere in the middle. So you have more control as in that you can develop a model that doesn't come out of the box with cognitive services like a completely custom model. But you still have some limitations, things like a 10 gigabyte data limit for instance, right.
You can still use custom code but you have to drag drop that as blocks that execute in a bigger model but here is the advantage, it lets you develop custom models and there is no local installation required. All you need is a modern browser and Azure subscription for deployment et cetera. In fact, you don't even need an Azure subscription to start with Azure ML, like learning it, but yes, in the real world you will need an Azure subscription.
Now when we talk about AI models, there are various steps we go through. Now let's talk this in the realm of Azure ML. You create a workspace, then you upload some data into there or import data from various places, then you create an experiment. So this is where you drag drop and create what looks like a flow cart. And then, you would train that experiment, now I'm over simplifying this picture, you train it and as a data scientist you would go through multiple training iterations to get the best possible results.
You may also have to retrain to improve your model even though the actual model may not have changed, right? So, training, retraining happens throughout the course of the model and then eventually how do the users consume it, you deploy it as a predictive web service. So this is the model, basically, you would send inputs into it, any program, and then they would get a prediction out of it, right? So you deploy that and as a web service and then various other consumers can use this web service.
Certainly, there are loops in this process like upload data, you may want to clean the data first. Training generally involves retraining, creating and experiment requires trial and error to identify the best algorithm that may suite for the current problem. So but as an overall high level process, this is what we generally do. Now this is not specific to just Azure ML, this maps very closely to any AI project but this is generally what we do.
But the important and interesting thing about Azure ML is that everything that you need to do can be done through your browser by going to https://studio.azureml.net. So, in this course I'm not going to go into the details of say, creating a model and various AI concepts around that. There are other courses on LinkedIn that I have that cover that fairly well. However, in this course we will just create a simple model and then we'll learn things like managing and deploying this model, you know the lifecycle of it, that's what I'll be focusing on.
But before I can talk about managing it, we do need a model so let's dive into studio.azureml.net and create ourselves a simple experiment.
- Creating a machine learning workspace
- Creating and training an experiment
- Creating a predictive experiment
- Deploying an experiment as a web service
- Enabling logging
- Viewing logs for diagnostic purposes
- Scale and geographic deployment of your service
- Using machine learning with API management