Machine learning? Keep it simple

Share this post:

When it comes to productivity the human mind doesn’t like to be distracted. Distractions are disruptions from deep work. And that’s precisely the problem with being a data scientist. See, your datasets have a story to tell and as a data scientist it is your job to listen to that story carefully. If you want to hear, understand and analyse that story, you’ll need an uncluttered mind. Meaning: a comfortable working environment. You want a powerful platform, flexibility in applying your data model, and freedom in applying your own toolkit, all while having no distractions from technical issues such as maintenance, debugging, and connectivity.

Until now the challenge has been to address all these needs simultaneously. Many companies have created platforms that attempt this. However, most platforms still do not fully manage to make things… simpler. Whereas simplicity is key. First, because it saves you from the distractions mentioned above. Second, because it’s useless to create a complicated data science model that is time consuming to maintain when it delivers little result for your organization.

And with machine learning being an essential element in data science, this becomes even more obvious. It’s all about tuning and applying a model using parameters (automatically) learned from previous cycles. Deploying and using a machine learning model should make things simpler. Strangely enough we often tend to overlook that. I once got handed a project that required me to create a real time automated model and I found myself in a permanent loop. My approach has always been to define goals, prepare the customer data and possibly create new data by using feature engineering, then create a model that’s easy to understand for my colleagues and customers. After that I tried to enhance results and overall confidence by importing more data, only to discover that I had to tweak and redefine both the model and the project goals.

In machine learning, the model’s accuracy defines whether we can trust the insights it generates. We can improve this accuracy but should keep the model simple as well. Otherwise we keep going back to redefine it. And here’s where IBM’s DSX (Data Science Experience) makes life easy for me since it creates a social environment where I can easily collaborate with other users to solve data challenges with the best tools and latest expertise. It has every machine learning tool you need, it provides a graphical interface for creating models, and it has a programming interface. It not only makes things simpler, it also pushes me to take a more focused and goal oriented approach.

I’m a data scientist myself and I know how important it is to be able to start work immediately. I want to have my datasets instantly ready for use without having to worry about connectivity. Thus, in using machine learning to find optimizations that create real world impact, don’t underestimate the power of simplification.

Take a look at IBM Data Science Experience.