Introduction
Overview of items covered in this post:
- Azure ML experiment workflow process.
- Overview of Data manipulation.
- Overview of Feature modules.
- Overview of training and evaluating model.
Azure ML is workbench software and it based on the workflow process to develop models and train it. Azure experiments are based on workflow depends on data required cleaning and training. The processes are as below:
Azure ML provides various ways of making data available for experiments. Data is a source to define algorithms and learn about an object.
We can learn about some of the data modules in Azure ML.
Enter Data Manually Module
Enter data manually module use to enter data manually instead of using any data source. It can be used for small data collections.
In most cases, we don’t use this module, but when we need a sort amount of data for an experiment, it is useful.
Import Data Module
It’s the most used method to import data in Azure ML studio and has a wizard way to import data from other sources. It is basically used to available external data in Azure ML studio.
![]()
You can select any of these data types to load data sources. It covers a vast area for capturing data from any available sources.
Unpacked Zip dataset
Unpacked zip dataset as per name it is quite clear it’s a compressed data source. When a large dataset needed for experiments, you can zip up and upload unpack and make it available in the data source. This method is quite useful because we play for large data and it reduces time to upload which is helpful in most scenarios.
After data is available in Azure ML, the next step is to have valid data for your experiment. Most data has missing rows and columns. Azure ML provides a very useful module, Clean Missing Data. It has various features like replace missing data using MICE, Replace using Mean, Replace using Median, Replace using Mode or Remove using row and columns. These are quite technical things which I will write about in other blogs.