The First Rule of Machine Learning: Start Without Machine Learning (eugeneyan.com)

Article summary

The article discusses the importance of starting without machine learning when approaching a new problem. It suggests that a non-ML baseline should be established first, using simple heuristics and data analysis, before considering machine learning solutions. This approach can help ensure that the problem is well understood and that the data is of high quality. By starting without machine learning, developers can create a more effective and maintainable solution.

Main themes

• machine learning
• software development
• legacy systems
• innovation
• career advancement
• technical debt
• programming languages
• system maintenance

What commenters say

• Some developers believe that using outdated programming languages like VB6 is acceptable as long as the code is maintained and works well.
• Others argue that using deprecated platforms like VB6 is a sign of a poorly maintained project.
• The need for constant change and innovation in software development can lead to unnecessary complexity and feature bloat.
• Some commenters think that the desire for career advancement and resume building can drive unnecessary changes to existing systems.
• There is a debate about whether replacing legacy systems with newer solutions is always necessary or beneficial.
• A few commenters suggest that the pursuit of modernity and fashion in software development can be detrimental to the overall quality and stability of the system.
• The importance of understanding the problem and data before applying machine learning solutions is a key takeaway from the article.
• Some developers feel that certain programming languages, like Java, are unfairly criticized and can be well-suited for production environments.