In Focus

Algorithms Of Microsoft Azure Machine Learning

Every Data Science and Machine Learning beginner has only one question in mind: how to start with Machine Learning or how to choose algorithms in Machine Learning, So let's understand Algorithms of Microsoft Azure Machine learning in this article

Every Data Science and Machine Learning beginner has only one question: how to start with Machine Learning or how to choose algorithms in Machine Learning. So let's understand Algorithms of Microsoft Azure Machine Learning in this article. Broadly there are 3 types of Machine Learning algorithms,

  1. Supervised Learning
  2. Unsupervised Learning
  3. Reinforcement Learning

In Supervised Learning algorithms in Machine Learning they are based on a set of particular examples; for example, some historic data like stock value and temperature can be used as guesses for future values. It also looks for different types of patterns like relevant data of that stock value; for example, company sector, date, time, past record, companies some account financial data etc and find correct pattern and using that pattern it makes prediction for unlabelled testing data and predict the future value. Supervised Learning is a very popular and famous machine learning algorithm type like Classification, regression and anomaly detection.

Classification means when the data is used to predict some category, you need to predict whether the image is of a Cat or Dog when you have only two choices for prediction the prediction is two class classification and when you want to predict the winner of ICC Championship Trophy that is an example of multi class classification.

Regression algorithms mean you are going to predict some future values like tomorrow's temperature or stock values etc.

Anomaly Detection means the goal is to identify data points that are simply unusual like in fraud detection suddenly you are getting an unusual credit card transaction from a remote location and an unusual spending limit is suspect. So in anomaly detection algorithms we simply learn what normal activities look like and identify anything unusual and significantly different also.

In Unsupervised Learning, Data have no Labels associated with them so unsupervised learning algorithm's goal is to organize the data in some way or to describe and identify the structure of data. So the task with this type of algorithm is to group them into clusters or find some ways of looking at complex data in simpler and more organized ways.

In Reinforcement Learning, the algorithms choose and decide the next action, as with robotics after some input the robot makes the decision to take next actions and periodically improves also, but in Azure ML studio right now there is no Reinforcement algorithm available. But these types of algorithms are required in the Internet of Things.

What we need to consider before choosing algorithms,

Even experienced data scientists compare algorithms before choosing the final one. We need to keep in mind the following points before choosing particular algorithms,

  • Accuracy
    It's not possible to always get an accurate answer with probability theory. Sometimes approximation is enough.

  • Training Time
    The time it takes to train the models varies a great deal between algorithms.

  • Linearity
    Many Machine Learning algorithms use linearity.

  • No. of parameters
    Number of parameters are the knobs when data scientists are setting them up  in algorithms. Sometimes unnecessary parameters also waste time, so sometimes initially it takes time to set everything up.