Modern Data Science platforms offer a gigantic set of features that usually cover everything you need for a Data Science project. And if you compare programming code and the visual process-flow based approach of these platforms, you immediately see the superiority of the visual approach: You see a structure within that even as an unexperienced user within seconds. Why is that? Because the 2D layout of a flow utilizes the human's automatic visual pattern finding capabilities that we have trained our entire life. In contrast to that programming code only "misuses" the eyes to read single characters that we actively need to interpret to form a program that we actively need to interpret to understand what's going on.
Only the complexity in this explanation should make clear: The visual approach in data science is a must for several reasons. For one it drastically reduces the costs for maintenance as one understands what's done much faster without seeing every detail. Second and even more important is that Data Science is a very interconnected topic: Data Science not only needs to talk to IT, but also to business holders. The visualization of what's being done makes it much easier to communicate it.
What do we need development for?
This said, it's pretty clear that our development is not designing solutions for entire projects in some obscure analytical scripting language. It is just here to support what really matters: A smooth execution of Data Science projects.
Sometimes Data Scientists face very specific problems that are so rare that the platforms don't support a direct solutions. Ever encountered a binary sensor data stream or something even more bizarre like time stamps in the French revolutionary calendar developed under Napoleon? Being faced with the decision how to go on, you usually have only two choices: Give up or break the tool chain and have a quirky workaround relying not only on a different product, but probably also on a specific system with a certain operating system having that product installed properly. Two years later, e.g. during a server transfer on a more powerful hardware, this can surface again as a really surprising problem, deeply hidden behind a thick layer of new business processes spanning over that...
How we will help
Our Development offers a clean and maintainable solution: We integrate the required functionality directly into the used platform utilizing their extension or plugin interfaces. That way, the functionality becomes part of the platform, is easily maintained from the Data Scientists perspective as it's now visually part of the data flow. At the same time it removes dependencies for the system administrator and avoids problems due to side effects in any change.
If this is not an entirely customer specific problem, we will publish the solution into a publicly available extension for this platform. If there are other customers facing the same or a similar problem, these extensions form a solid basis and easily can be extended for covering related use cases reducing the development costs for the single customer. Even more, customers benefit from the updates of the extensions and with the additional features.
For the last year, we have been using the Statistics Extension [...]. We've been very pleased with Old World Computing and how they have incorporated our customer feedback into subsequent releases of their tools.
Brian Tvenstrup - Chief Analytics Officer - Modern Marketing Concepts Inc.