8 Aug 2019 · 9 mins
I started my MSc Business Analytics course at the University of Surrey almost one year ago. I had no prior experience in Machine Learning or data science. Before, I used to develop and manage EU projects for businesses, local authorities and non-profit organisations. I even achieved two International awards for best project. However, I wanted to immerse in the technology field and be part of the great community which enhance business and people’s quality of life by developing the products and services of the future. Therefore, I decided to undertake a new adventure in my life and to pursue an entirely new career path. Yeah, I am a “game-changer”. ☺
Through the MSc Business Analytics course, I was hoping to equip myself with the skills and knowledge necessary to start a promising career in the field of data science. At university, we learned to use software for advanced analytics like EViews, SPSS Modeler, SPSS Statistics, R and RStudio. They enabled me to begin building machine learning models and apps to gain insights from data for effective business decisions. However, a lot of the tasks had to be done manually, which takes time. For example, tuning the hyperparameters of a model, or building different assessment graphs, as well as the comparison of models, and determination of the best performing model. Also, coding an entirely new machine learning application from scratch can be time-consuming.
In July 2019, I started my internship at Boemska. Thanks to them, I started exploring the world of SAS. SAS® Viya® “shocked” me in the most positive way! It is amazing! I found out that in the field of software solutions for Machine learning, SAS Viya is a “game-changer” ☺. Here is why:
SAS Viya is a cloud-enabled, in-memory analytics platform that supports faster processing for huge amounts of data and complex analytics; a standardised code base that supports programming in SAS as well as open-source; and deployment for cloud, on-site or hybrid environments. The features of the platform make the collaboration and use of advanced analytics tools more accessible to everyone in the enterprise - data scientists, business analysts, developers and decision-makers. SAS Viya hosts all necessary solutions for the execution of the entire analytics life cycle. You can access them through SAS Drive, which is a web-based single point of access. In this blog post, I wanted to focus mainly on the exploration of the characteristics and advantages of the SAS® Visual Data Mining and Machine Learning solution.
SAS® Visual Data Mining and Machine Learning consist of data mining tools which enable you to build and put into production predictive models. The access to its features is enabled by Model Studio – a web-based visual interface.
What are the Benefits of SAS Visual Data Mining and Machine Learning? Once you open Model Studio, you instantly start bumping into its advantages. Here, I share my thoughts on the ones that impressed me the most.
To help you get a better understanding of the topics presented in the article, I am using screenshots from SAS Viya for Learners. This excellent cloud-based software suite can be used for free by educators and learners (Khan, 2019). I will write a follow up about SAS learning in a separate article. Now let’s have a closer look at Model Studio.
SAS Viya has a modern and at the same time simple interface. The first thing you notice is that all your projects are well organised in one place, in SAS Drive. You create or open your Model Studio projects from here.
In every project, there is a data source where you can easily switch between two views:
1) to explore all variables
2) and to view the data table:
The coolest feature is that you can view at a glance the whole well-structured analytical workflow in a single interactive diagram, called “pipeline”. Each analytical task is represented by a node in the pipeline.
You can just drag and drop a node from the Nodes pane (on the left part of the screen above) into the pipeline to include a new analytical task in your Machine Learning predictive model. ☺ That makes the navigation of the whole workflow while building a data mining model ridiculously easy and quick. An expert with limited technical skills could master the workflow only one day after they start using Model Studio. Moreover, you can build as many pipelines and models as you need in your project and quickly navigate across them.
Once you add a node (analytical task) to your workflow, in minutes you can set up its parameters in the properties pane which is at the right side on the same screen where your pipeline is:
This feature helps you to find the optimal hyperparameters of a Machine Learning algorithm you use to build a model in your pipeline. Isn’t that awesome! The autotune feature saves you lots of time when you search for the best settings to create the best possible model.
Once you have built your model, Model studio automatically creates many interactive beautiful and comprehensive assessment visualisations in graphs and tables. They provide many fit statistics which you can use as a benchmark to assess and compare models’ performance. As you can see below, they are presented in a well-organised manner:
This automation saves you time and helps you to focus on the important part of your job: to quickly, easily and reliably assess the performance of your models.
Model Studio automatically chooses the best model among all models you have built and tested in your project. In a single pipeline, Model studio automatically adds a ‘Model Comparison’ node:
It compares the performance of all models in the pipeline, ranks them upon various criteria and chooses the champion model:
Then, Model studio automatically chooses the champion of the champion models across all pipelines in your project:
You have built an awesome predictive model in a pipeline, and you think it could be used as a template by your colleagues in future projects? Alternatively, you might be looking for a ready to use recommended node template for a specific analytical task? The Exchange feature enables you to quickly and easily create and share node or pipeline templates with your team. You can filter your search in the Exchange repository by type of Machine learning project (e.g. forecasting, text analytics) or type of analytical task (e.g. preprocessing, modelling). This cooperation accelerates your work and boosts your productivity.
So, if you already think that building predictive Machine Learning projects in SAS Viya is fantastic, let me show you some extra advanced features of Model Studio. You will fall in love with SAS Viya, as I did. ☺
Some extra awesome features:
Machine Learning algorithms like the Support Vector Machine (SVM) are widely considered as “black boxes” as they are almost impossible to interpret. In many cases, they are valued for being able to predict a highly accurate outcome from a model. However, many practitioners cannot use them to explain the relationship between the input variables and the target variable.
Model Studio’s automated model interpretability feature changes this. It enables even experts with less experience and technical skills to quickly and easily find some meaningful insight into the relationships between the predictors and the target variable in “black boxes” as SVM.
When you select the automated interpretability feature, it analyses the resulting predictions of the model and creates several interactive model interpretability plots: The Partial Dependence (PD) plot, the Individual Conditional Expectation (ICE) plot, and the Local Interpretable Mode-Agnostic Explanations (LIME) plot. They visualise the relationship between the input variables and the target variable. Besides, the Input Relative Importance table presents you with the most important predictors in the model.
Thus, the “Game Changer” SAS Viya transforms the “black box” of the SVM algorithm into a comprehensible and interpretable model which quickly provide an enterprise with valuable business insights.
Moreover, you can use the model interpretability feature to compare different models from different Machine Learning algorithms. This opportunity additionally increase the transparency in choosing the best model of your project.
If only my coursemates from the university could see this! ☺
Model studio enables you to seamlessly integrate different advanced analytics tasks in the pipeline while developing your models. For example, you can easily combine and add both structured data and unstructured data to your model. With the text mining node, you transform unstructured text variables into numeric predictors in minutes:
Once you have chosen the champion model of your analytics project, you can put it into production in any environment with just a few clicks in the SAS Model Manager. SAS Viya automatically translates the model into score code which then can be run on a new dataset.
SAS Viya enables you to access your code through the most popular languages for data analytics - SAS, Python, R or Lua. This integration gives you more freedom and flexibility to manage your analytical projects using consistent and governed SAS code. Also, you do not need to write any Machine Learning code from scratch, which saves lots of time.
Moreover, you can use SAS Viya solutions through your own preferred interface. Tools like Boemska’s AppFactory empower your Citizen SAS Developers and digital/UX teams to easily build modern data-driven App interfaces using the same SAS Viya infrastructure. This way, you can quickly have a new tailored interface for the SAS Viya solutions which account for your specific business processes needs. Having your own preferred interface means that your team feel more comfortable in using the SAS Viya solutions, and you can accelerate your Machine Learning productivity.
I can conclude, being at the beginning of my journey in the world of Machine learning, that SAS Viya is a game-changer, which inspires and motivates me to immerse more and more in the data science. It makes it so much easier to build predictive models. Moreover, in SAS Viya environment, I can do it faster. The intuitive navigation, guided and well-structured visual analytics workflow, alongside with the automated features and the easy to set up parameters of the analytical tasks democratise machine learning and make it accessible for everyone in the enterprise. At the same time, the sophisticated and seamless integration of many models, team collaboration and project governance boost the Machine Learning productivity of your team.
To find out more about SAS Viya, have a look at the SAS website here.