In this video, explore the history of predictive Analytics Workbenches.
- So why use an Analytics Workbench? Most analysts and their managers divide advanced analytics choices into just two buckets, tools with graphical user interfaces and coding, notably of course R and Python. And survey show that most analysts do indeed learn some R or Python along the way. But I think that some, especially if they're new and unestablished, are afraid that unless they write raw code, that they won't be taken seriously. So given the choice between easy and hard, why is the more difficult choice so popular? Well, there's good reasons to favor code, the idea's that code is going to be more deployment-friendly and it will more completely document your work. Also the feeling is that anyone that overcomes the challenge of learning R or Python will also be a good modeler. While this is often true it's not always true. Some code is badly documented and unfortunately, not all coders are good analysts. So what's the solution? Well, there is good reason that some are worried about fancy graphical user interfaces, they're often associated with analytics tools that are largely automated and that can be used by business users that haven't learned about data science and predictive analytics. The marketplace is still sorting out the role of these tools and vendors are still working to make automated machine learning a reality. We'll leave that debate for another time. However, graphical user interface versus coding is a false dichotomy, there are numerous options and for many an Analytics Workbench is a great solution. It offers what is often called visual programming. You can rapidly prototype, by drawing what is essentially a flowchart, but it's completely customizable. These are tools that workees can acclimate to with some training, but are powerful enough for true experts, they offer a middle ground between fully automated and raw code, which makes working in them faster and easier, especially for routine tasks. In fact, routine data prep was the inspiration for the first workbench-style interface, which was developed almost 30 years ago. Colin Shearer, who designed the very first workbench, described it this way, "We were finding that data mining projects involved a lot of hard work, and that most of that work was boring. Unearthing significant patterns and delivering accurate predictions, that part was fun, but most of our effort went on mundane tasks, such as manipulating data into the formats required by the various modules and algorithms we applied". So that's why an option like KNIME makes so much sense. It eliminates or at least speeds up the boring tasks, yet it's essentially self-documenting and because the flowchart is easy for anyone to read when it comes time to deploy you've got a well-documented production-ready process. And if you ever have a need that isn't met with a KNIME itself, you can incorporate any raw code that you like, including R, Python and many other languages.
- Why use a workbench
- Why choose KNIME?
- Adding KNIME nodes with extensions
- Accessing data
- Exploring data statistically and visually
- Merging and aggregating data in KNIME
- Modeling in KNIME
- Scoring new data
- Combining KNIME with R and Python