Monday, October 13, 2014

Data Science/Data Scientist...

With the advent of Big Data technologies, the increased use of non structured data in business analytics, one of the roles that is beginning to rise in terms of demand is the Data Scientist. The area of study that a Data Scientist specializes is the area of Data Science. The huge amounts of data has called for more predictive analysis of the data instead of just focusing on the operational data. The power of predictive analytics if done right is that it gives the ability to a business to make better decision about future plans. For example how a company should target is products, what type of features does a product need to have to address the customer requirements. The emergence of data science has made Microsoft re prioritize the data mining features they incorporated into SSAS. All of the Data Mining algorithms are being re purposed into the Machine Learning tool in Microsoft Azure. I have explained about the Machine learning topic in my previous blog topic. In addition to Machine learning, to be make more effective usage of the tools available in Azure, it is important to understand the basics behind Machine Learning. Coursera offers a course on R Programming, it is free of cost and one can enroll by going to the website. In case one is interested in going beyond learning the R language, Coursera has a complete Capstone course in Data science which includes learning about the algorithms behind data mining, getting introduced to machine learning apart from learning the R language. The Capstone course covers experimental design, statistical inference and regression models. This is one area which is growing very fast, so having a good foundation with courses listed above can provide a step into the predictive analytics world. One can also refer to the following link to understand the different types of algorithms in Data mining.

Data Science Capstone Course link:

Monday, October 6, 2014

Hive, BI Best Practices, Machine Learning...

The SQL Saturday BI edition that was held in Charlotte on October 4, 2014 had some interesting topics which I attended. They were really insightful and provided me with some good information that I would like to share. The first session by was given by Kevin Goode on topic of Introduction to Hive, the session was really interesting with lots of questions and Answers. HIVE basically is a layer on top of hadoop that allows users to write SQL language type queries. HIVE in turn take the SQL queries and converts them to map reduce jobs, they get executed the hadoop layer. The results are then provided back to the user. HIVE uses a language called HQL which is a sqlvariant and has lots of similarities like MySQL and Oracle Syntax. The query construct of HQL has commands which very similar to Transact SQL. One of the main things that need to keep in my mind is that HQL/HIVE does not great optimizer like sql server, so while constructing the join statements in HQL one has to be very careful about placing the tables in the join. HIVE?HQL is best suited for anaylytical queries and also it has no updates and deletes. The rule of thumb for now seems to be that the smaller tables come first and then the larger tables. HQL/HIVE also provided some type of execution plan but these are not as sophisticated as the ones you see in SQL Server.  The error messages provided by HQL when there are issues are more java based, so it could a little bit of getting used while working with HQL error messages. While working with HIVE/HQL the developer should be aware of the type of data coming in and how it is partitioned. The organization of data really helps optimizing the HQL queries. The results provided by HQL can be used to feed a traditional datawrehouse system.One of the areas where Hadoop is weak is in the area of security. Security is not very structured in hadoop, the security has to be achieved using other bolt on tools. Hortenworks is one of the leading distributors of hadoop and they provide information regarding HQL. The frontend provided to connect to HIVE is completely web based and users can use this tool to write HIVE queries using HQL. Hortenworks is one of the big players in hadoop distributions: please visit the hortenworks website here.
The second session which attended was on tabular models. The topic was about how build a tabular model using visual studio 2012 with SSAS 2012 tabular instance installed. It was an informative topic as it bought out the differences between SSAS traditional MDX based and tabular solutions. One of the key differentiators is that SSAS is disk based and tabular is in memory and provide good speeds. The limitation of the tabular model would be size of memory that can be used to hold tabular models. Also with tabular you don't have writeback capabilities. The demo was pretty impressive, some of the steps done in the demo very similar to how one would go about building SSAS cubes.  The topic was presented by Bill Anton and his blog is located at:
The third session which attended was the most crowded of all the sessions. It was about Best Practices for delivering BI solutions. The audience were all BI professionals and also were with IT in various companies. SQL Server BI Expert James Serra provided a great session on best practices. He started off with why BI projects fail, a Gartner study has revealed that around 70% of BI projects fail. The fail factors ranges from lack of expertise/experience, lack of proper communication, requirements and project management. He highlighted the issues of clash between Business and IT over how the solution should be delivered. One of the interesting aspects that was mentioned was for technology to provide a clean data ware house and then build some abstraction layers on top of the warehouse, once that is done allow the users to build the users to utilize self-service BI solutions.Understand the difference between kimball and Inmon priciples, cappabilities of tabular vs Multidimensional, star schema vs relational. Please visit his website for some great information on this topic and other aspects related to Datawarehouse.
The final session I attended was about Introduction to Microsoft Azure Machine Learning, this topic was really exciting and informative. The topic is basically about predictive analytics, the tools provided by Microsoft as part of the Azure framework. Microsoft has a group called Microsoft research which initially working on developing algorithms for the Data mining product as part of SSAS. In the past few years there has not been much push on data mining within SSAS. Recently with Microsoft making a big push for cloud services offerings as part of the Azure framework, all the Microsoft research are now being made part of Machine learning using Microsoft Azure. He is a link that provided a head start on the Machine learning topic: During the Session Bill Carroll provided an insightful demo of the machine learning framework within Microsoft Azure.  Azure provides a very nice framework for a developer to set up an experiment. The workspace for setting up experiments look very similar to SSIS workspace within visual studio. As part of the setting up experiment, one feeds data to an algorithm based on the type of outcome that is needed, then the model is trained, scored and evaluated, the model can be published. Once the model is published, it can be called through an application using a web service. There are code snippets available in C#, Python to call the web service to execute the model. Some of the examples of machine learning are when a user purchases a product in amazon, there are a list of recommended products that are generated, list of recommended movies in Netflix based on users watching preference, and these are some of real life examples, where Machine learning is used.