Microsoft Azure Machine Learning Review
Microsoft Azure Machine Learning Review. Hot on the heels of our Amazon Machine Learning Review, we decided to do a review and compare against Microsoft Azure’s offering of Machine Learning services on the cloud. In short, we find Microsoft Azure Machine Learning services quite amazing and liked it better than Amazon’s Machine Learning services.
Free Trial Access
Nothing beats free trial access to test drive the system the system for free. Woo hoo! Free $200 credits remaining. I am beginning to sing “…This used to be my playground. This used to be my childhood dreams…”
But we have to get real and check the real production pricing which can be found at Azure Machine Learning Pricing and copied below. The pricing model is unique. Base fee is low and hourly experiment fee is also low. I guess the bulk of the cost would show up if one publishes the API for production use and there is substantial transactional load on those API calls.
Azure Marketplace vs Machine Learning Studio
First thing I had to realize was how rich the Azure Machine Learning offerings were. There are pre-packaged working AI models available in the Azure MarketPlace For Machine Learning and there is Azure Machine Learning Studio where one can configure Machine Learning models using graphical interfaces. These are very different toolsets for different use cases.
For example Azure MarketPlace has packaged services like the following, some of which might have been built using Azure Machine Learning Studio and many many more.
Customer Churn Prediction
Customer Churn Prediction is a churn analytics service built with Azure Machine Learning. It’s designed to predict the likelyhood of a customer (player, subscriber, user, etc.) ending his or her relationship with a company or service.
Text Analytics API is a suite of text analytics services built with Azure Machine Learning. Just bring your unstructured text (English only), and use this API to perform sentiment analysis and key phrase extraction.
Recommendations API by Azure Machine Learning helps your customer discover items in your catalog. Customer activity on your website is used to recommend items and to improve conversion in your digital or physical store.
Frequently Bought Together
Frequently Bought Together is a market basket analysis API built with Azure Machine Learning. It helps your customers discover items in your catalog that are frequently purchased together. Use your purchase history to add recommendations to your website.
Binary Classifier API
Binary Classifier API is an example built with Microsoft Azure Machine Learning that fits a logistic regression model to user inputted data and outputs the predicted value for each of the observations in the data. Suppose you have a dataset and would like to predict a binary dependent variable based on the independent variables. ‘Logistic Regression’ is a popular statistical technique used for such predictions.
You can subscribe to a service in the marketplace and try to use it. For example, here we try the text analysis service.
It was very simple to set up. However, I was wondering why the supported languages didn’t include usual internet languages like Python and Ruby.
Experiment with Text Analytics
We went ahead and tried exploring the text analytics API in the browser without code.
We took some text from a Marketwatch article:
Text 1: As much as the market rallied this past week, we still have not seen dissipation of the bearishness in the financial media. Most seem way too focused on corporate earnings, and believe that will be the next shoe to drop, and cause the market to drop.
Key phrases identified were: “bearishness”,”financial media”,”market”,”past week”,”corporate earnings”,”dissipation”,”shoe”
Sentiment score was: 0.287 (I believe where 0 is very negative sentiment is 1 is very positive sentiment)
Text 2: The upcoming week should provide us with the answer to the question I posed in the title of this update. Keep in mind we have been looking at the last month as a consolidation setting us up for a strong rally toward 2200 in the S&P 500 and 132-136 in the IWM. So, if next week does not provide us with weakness in the equity market, and we take out the all-time highs, we are clearly heading much higher, and toward our long-time targets much sooner than later. But should the market be so kind as to provide us with one more pullback or even one more low toward 2021ES, it will likely be your last opportunity to enter for this next rally to 2182ES/2189SPX and 132-136 in the IWM.
Key phrases identified were: “strong rally”,”weakness”,”long-time targets”,”title”,”equity market”,”us”,”question”,”time highs”,”consolidation”,”answer”,”kind”,”upcoming week”,”SPX”,”update”,”pullback”,”month”,”opportunity”,”mind”,”IWM”
Sentiment score was: 0.99 (I wonder why the score was so high?)
Text 3: Oh no. The world is coming to a sad end.
Sentiment score was: 0.00 (Ok, this makes sense)
Text 4: Oh yes. The world is a beautiful place.
Sentiment score was: 0.88 (This makes sense too)
Text 5: Oh no. The world is a beautiful place.
Sentiment score was: 0.55 (Now, I am just fooling around and confusing everyone)
Overall, the Text Analytics service performed as I expected and is slightly better than other Text Analytics engines I have tried.
Azure Machine Learning Studio
Now, Azure Machine Learning Studio is what makes the whole experience rock. The graphical user interface is powerful and yet amazingly simple and intuitive to use. First you create a workspace like following
You can find plenty of documentations at:
What is amazing is the array of samples and templates you can use for experiments eg binary classification, regression, credit risk anomaly detection, customer relationship prediction, flight delay prediction, prediction of student performance, twitter sentiment analysis, recognition of hand-writing, neural networks, online fraud detection, movie and production recommendations, time series forecasting etc.
You can see the full listing at the Machine Learning Gallery http://gallery.azureml.net/
Compared to the narrow set of possibilities at Amazon, this is where Azure really shines. Now I am wondering why I discovered Azure Machine Learning so late. Was it because I overlooked it in favour of open-source source libraries or typically open-source friendly vendors like Amazon and neglected to see how powerful Microsoft platforms can be?
Amazing UI in Azure Machine Learning Studio
The graphical UI in the Machine Learning Studio is wonderful to say the least. Feels like a sophisticated thick client but all in my Chrome Browser with animation, graphical flow charts to help you understand a complicated process.
I decided not to go easy on Azure and tried stress testing by running multiple concurrent model evaluations.
It does take quite a while to run some of the models. Filter Based Feature Selection was stuck for over 30 minutes for the Binary Classification of Twitter sentiment. The other experiments each only took about 5-10 minutes to complete which was acceptable.
The simple ability to share your results (eg on Twitter) is a nice friendly aspect great for their marketing too.
We have tried many machine learning platforms before from the powerful open-source ones like Python Sklearn, Mahout to open-source commercial ones like 0xData, Prediction.IO to large vendor on the cloud ones like Amazon Machine Learning and now Microsoft Azure.
In terms of being able to write code in however way you like and integrate easily into your code, Python Sklearn is probably still the most flexible but comes with a medium learning curve and only for developers. But if you want a Machine Learning service on the cloud for beginners to experts to even business users, with a powerful and intuitive user interface at a decent cost, I think Microsoft Azure Machine Learning Studio and Marketplace clearly outshines all other Machine Learning solutions on the cloud.
If you like our articles, please follow and like our Facebook page where we regularly share interesting posts.