Math Skills vs. Pragmatism in Data Science

Math skills are often considered fundamental to becoming a data scientist. This is true, but we often forget to mention that in practice, machine learning is above all a pragmatic discipline.

It's important to stress this because theory and practice sometimes pursue distinct objectives, and some students who love math could be disappointed by this on-the-ground reality.

This pragmatism means that we don't judge an idea by whether it's beautiful or neatly constructed, but by its practical consequences. In other words, whether it yields good results, which can be frustrating at times.

Even though math is fundamental, data science in practice is not the pursuit of pure truth.

It's not the admiration of an elegant formula that solves everything. It's not about finding the right way to converge towards the global minimum (if that's even possible). Large language models themselves are the result of an idea that seems very simple a priori: predict the next word based on the previous ones. There is no explicit mechanism that would allow it to reason. But implemented on a massive scale of data, it produces impressive results. If this approach has become widespread, it's not because it was the most elegant idea, but because it's the one that works best, for now.

The data scientist must be prepared:

• To examine the data in detail, even if it's laborious
• To be able to abandon appealing but ineffective ideas
• To explore avenues, however simple or counterintuitive they may seem at first
• To sometimes focus more on engineering at the expense of the scientific aspect

I'm not minimizing the value of mathematical theory in the least - it remains essential. I'm simply offering a more nuanced perspective. Because in data science, the value created is measured above all by the ability to solve client problems or create opportunities.

If we can combine efficiency and elegance, so much the better. If not, we'll keep just the efficiency.