Beginning with this article, I am going to start writing a new series on Machine Learning using Azure Machine Learning Studio. Microsoft launched Azure Machine Learning Studio last year, for data analysis, predictive analysis, data mining, and data classification etc. Microsoft has already implemented most of the classic machine learning algorithms in Azure Machine Learning Studio. We don’t need to write our own data analysis algorithm if we use Azure ML Studio but still, there is opportunity for us to design our own data analysis algorithm using Python and R language.
This article is going to be an introduction of Azure ML Studio where I will talk about machine learning, Azure Machine Learning Studio subscription, data type, algorithm categories, and algorithms. Let’s talk about Azure Machine Learning Studio.
Machine learning is a way to understand the data pattern, recognize it, analyze it, get insight of it, and predict accordingly for the future. It helps in data mining, language processing, image recognition, and Artificial Intelligence.
We can split machine learning into several areas - supervised learning, unsupervised learning, and recurrent learning. In this tutorial series, we will only talk about supervised and unsupervised learning.
Supervised learning is predicting the future by learning from past examples.
Unsupervised learning is understanding the past, making sense of data, learning the structure of data, and compressing data for consumption.
Azure Machine Learning Studio
Azure Machine Learning Studio is built considering the cloud-first mobile-first world.
- This tool is fully managed and we don’t have a need to install any external software, or manage any hardware, as there is one portal to view and update everything.
- It’s very simple to use. Simple drag, drop and connect interfaces are there to use for data science features. We don’t need any programming for common tasks.
- 33+ classic and mostly used data analysis algorithms are already implemented on this tool. We don’t need to implement new algorithm for basic use. But, it supports R and Python for extensibility usage.
- We can create service and integrate our data analysis tool with our existing web, mobile, or desktop application very easily. We can monetize our designed data analysis model in machine learning marketplace as well.
- We can collaborate work with anyone, anywhere via Azure workspace.
- There is visual composition with end2end support for data science workflow.
- There are built in immutable libraries of models. So, developers just need to search, discover, and reuse them according to their need.
- As we don’t need to do implement our own code, we can test our dataset with different algorithms and models in few minutes.
Azure Machine Learning has endless possibilities. We can perform social network analysis, predictive maintenance, web app optimization, healthcare outcomes, natural resource exploration, fraud detection, churn analysis, life science research, network intrusion detection and many more.
For any type of data analysis application, at first we need proper dataset. To develop a data analysis application for Azure Machine Learning Studio , we need to setup cloud environment. Azure Machine Learning Studio is there for us, it is integrated with our Azure cloud environment and it is fast, efficient, and very easy to use.
Azure Machine Learning Studio supports different types of data format. It supports,
- ZIP as data format.
We can use different data sources in Azure ML Studio. Some of the supported data sources are,
- Azure Blob Storage
- Azure SQL DB
- Azure SQL DW
- Azure Table
- Hadoop Hive Query
- OData Feed
- Web URL(HTTP)
- Desktop Direct Upload etc.
We can use different types of data preprocessing methods to explore and process our collected data. Some of them are,
- Clean Missing Data
- Clip Outliers
- Edit Metadata
- Different Filters
- Feature Selection
- Normalization of Data
- Quantization etc.
We can use different methods of data visualization in Azure Machine Learning Studio. These data visualization models are already implemented. We just need to integrate in our system and use them. Some of them are,
- Box Plots
- Bar Charts
- Python Plotting Libraries
- ROC Graph
- Decision Tree
- Confusion Matrix
On 8 major categories, 33+ classic and mostly used machine learning algorithms are already implemented in Azure Machine Learning Studio. Major categories and algorithms are,
- Anomaly Detection
- One-class support vector machine
- PCA-Based Anomaly Detection
- Time Series Anomaly Detection
- Bayesian Linear Regression
- Boosted Decision Tree Regression
- Decision Forest Regression
- Fast Forest Quantile Regression
- Linear Regression
- Neural Network Regression
- Ordinal Regression
- Poisson Regression
- Two-Class Classification
- Two-Class Averaged Perceptron
- Two-Class Bayes Point Machine
- Two-Class Boosted Decision Tree
- Two-Class Decision Forest
- Two-Class Decision Jungle
- Two-Class Locally-Deep Support Vector Machine
- Two-Class Logistic Regression
- Two-Class Neural Network
- Two-Class Support Vector Machine
- Multi-Class Classification
- Multiclass Decision Forest
- Multiclass Decision Jungle
- Multiclass Logistic Regression
- Multiclass Neural Network
- One-vs-All Multiclass
- Computer Vision
- Text Analytics
- Feature Hashing
- Named Entity Recognition
- Vowpal Wabbit
- Statistical Functions
- Descriptive Statistics
- Hypothesis Testing T-Test
- Linear Correlation
- Probability Function Evaluation
We don’t need any Azure Subscription, Credit card, to create Azure Machine Learning Studio account. We can chose anonymous guest access or sign in with our work, school or Microsoft account to use Azure ML Studio. But, if we have Azure subscription, then our Azure ML Studio account will automatically integrate with Azure account.
We can create Azure ML Studio account from this site, https-//studio.azureml.net/
Figure 1- Initial page of Azure Machine Learning Studio tool
From figure 1, we can see sign in and sign up option on https-//studio.azureml.net/
Figure 2- Azure Machine Learning Studio Dashboard
After signing up or signing in into the system, we can see the above dashboard page. Our existing experiments will show on the first page with a diagram at the right. On the left bar, we can see features and they are,
- Web Services
- Note Books
- Trend Models
Figure 3- New Experiment Selection
There is a section at the bottom and if we click on the +New button at the bottom, then Figure 3 will appear. Here, we can select experiment templates from the Experiment section. Project and Notebook are still in preview mode. Notebook is for Python Jupyter notebook. We can create blank experiment from here.
Figure 4- Settings Page
From the Settings section, we can see my workspace name, ID, and workspace storage. For each guest user, Azure ML allocate 10 GB of data to use for free.
Figure 5- Existing Datasets
If we click on to Datasets option, then we can see that lots of standard datasets are already there for us to test our system and to learn about different models.
Figure 6- Existing Implemented Experiments
If we click on to Experiments section, then we can see that lots of experimental models are already there for us. We can search our suitable model, discover them and reuse them.
Figure 7- Structure of Sample Model
From the above figure, we can see Sample Model 3. If we click on the existing experimental models, then we can see each model's diagram like below. This model is the combination of different components.
Figure 8- Visualizing existing model data
If we click on the cross validate model and then right click on this component, then we can see options like the above figure. Click on the Visualize option to see the data.
Figure 9- Data visualization of existing model
After clicking on to Visualize data, we can see section like above image. This is the raw data and this model is designed to analyze that raw data.
This is the first article of a long series of articles on Machine Learning and Azure Machine Learning Studio. It’s just an introduction article of this tool. From the next article, we will pick each algorithm, will describe the algorithm in very easy words, and then develop a machine learning application using that application from the scratch.