Let’s look at those steps:
1. First, you need some data. The machine learning tools allow you to import data and tools are built in for analysis of that data, so you can do some quick visualizations and profiling. Modules are also built in for data cleansing, such as removing missing values, removing or dealing with outliers, doing data normalization to bring things into range for certain models. If you’ve written your own R or Python scripts that do this, you can integrate those scripts into your machine learning model.
2. Next, the machine learning toolset allows you to choose a model. In broad terms, choosing from either a progression model, a classification model, a clustering model or a model for anomaly detection.
3. Now it’s time to test the model. There’s tooling available that allow you to split your dataset into 2 pieces. One that you’ll use to train your model and then after your model is trained, you use the other half to do testing. All these tools are available in the Azure Machine Learning toolset.
The process is different than other technical problems in that there’s no definitive goal that you’re searching for. By nature, it’s an iterative process; the machine learning tools make it easy for you to iterate, try different models and tune them and test them side by side to evaluate.
4. Once you’ve selected a model, it’s time to deploy it. With the Azure Machine Learning tools, you can deploy to a web service, then the web service can be integrated into an application. For example, if you had an application and you were accepting some user input, you could take that input and feed it into a machine learning model that would render a prediction on the fly.
With the web service, you could also use Azure Data Factory for batch style predicting by inputting a batch of data and have that output a batch of predictions that you could store in a data warehouse or some other reporting application.
Bottom line, this is a very powerful toolset and if you’re not familiar with any type of machine learning or predictive analysis, it’s a good way to get started. If you have questions about machine learning, the tooling or the process, we are your best resource. Click the link below or contact us, we’d love to help you incorporate machine learning in your business.