In this blog post we briefly look into azure machine learning. First let's start off by answering this question: What is machine learning?
According to Arthur Samuel, an American pioneer in the field of computer gaming and artificial intelligence, machine learning is a field of computer science that gives computer the ability to learn without being explicitly programmed.
It is a method of teaching computers to make prediction based on large volume of data. It is a branch of artificial intelligence which automatically improves programs using data. For example a machine learning system could be trained on email messages to learn to distinguish between spam and non-spam messages. After learning it can be used to classify new email messages into spam and non-spam folders.
Now let's see what is Azure Machine Learning and what it has to offer us for machine learning.
Azure Machine Learning also called Azure ML is Machine Learning as a service or platform as a service offering for predictive analytics. It is a data mining and data science and machine learning tool in the cloud, however it is different than most data science tools, because there is no coding involved.
Traditionally data science tools involve R, Python, or Mathlab, which require coding, not with Azure machine learning. It allows to rapidly prototype and easily build, test and compare experiment results by dragging and dropping approach. It also does a seamless integration with SQL, R or Python and you can mix and match so all of a sudden you can be using SQL, drop SQL, use a module and then all of sudden switch again and then go into R. You can also deploy these models automatically by contacting your models in cloud via rest APIs.
The machine learning algorithms in azure machine learning offers 25 algorithms built-in broken up into 5 separate categories of algorithm.
In order to setting up a machine learning experiment in azure you need the following steps:
First step is to prepare and define your data set.
Then, select columns in the dataset. You define the features which is the properties of something you are interested in. In your dataset each column is a feature. This step is very critical and important and requires experimentation and also knowledge about the problem you are going to solve.
Now that you have the data ready, it is time to train and test the model for accuracy. The process is to use data to train the model and then test the model to see how closely it is able to predict.
As mentioned earlier there are different algorithms available in 5 categories and you need to choose which algorithm to use. For instance, classification and regression are two types of supervised machine learning algorithms and classification is used to predict an answer from a defined set of values such as gender male or female. Regression is used to predict a number e.g. price.
Next step is to score the model to determine accuracy by comparing the prediction that it is made against the training data with the validation model data.
Training data for machine learning experiment is extremely important. You can use your meaningful dataset or alternatively there are more than 35 sample datasets is available out of the box. Some of them are Airport Codes, Bike Rental UCI, Book reviews from Amazon, Breast Cancer data.
Once you built your machine learning experiment, then you need to publish a predictive web service. This web service then can be called by code and used in things like Azure Function, Web Job, Web App or any other application that would need to implement your machine learning service.
Let's get into what Machine Learning Studio has to offer.
The Azure Machine Learning Studio is a visual designer for developing machine learning experiments. With a web based visual drag and drop UI is cross platform and enables you to connect different pieces for building up your experiment. In order to access Azure Machine Learning Studio click on this link.
Here is a screenshot of what the azure machine learning studio designer looks like:
Now let's talk about pricing. There is two parts to the pricing of Azure Machine Learning Studio.
First is a monthly subscription and then second is for usage. You're charged for $9.99 per month per seat so for each workspace you will be charged about $10 a month.
Secondly is your charge for usage. There is three parts to the usage, the first on is runtime. So you are charged $1 per one hour of experiment runtime. Every time you hit the run button on your experiment, you are going to see a timer tick on the top right hand corner of your screen. That is what you are being charged for. The second part is you are also charged for deployment calls. Anytime anyone calls your deployed web services, your rest APIs, for every 1000 API calls, you are charged about $0.10 to $0.50 depending which tier you are going with for your web service. Keep in mind that the first 1000 API calls are always free. The third usage is you are charged for storage of data. So basically whatever data you pull into Azure ML, you will be charged for that as well. For instance, Azure blob storage is one option that charges almost $0.02 per gigabyte per month.
In next article, I will walk you through Azure Machine Learning Studio in action.
Number of Views:1003