Thanks Bojan for highlighting the fact that powerful methods don't get "outdated" when you have all the shiny and parameter-heavy models around. Perhaps we should wait for your next posts, because this one did not help clarify where you've made success using XGBoost. Looking forward and kudos!
My experience matches almost everything you have said...with the slight exception of "even specialized methods like support vector machines can be a better fit". I don't think I've had a situation where an SVM was the best fit in almost 15 years - do you ever still find it to be useful? Maybe I just haven't been in that type of specialized situation. (I still use Random Forests / GLM / GAMs, although not as often as XGB)
Yes, I indeed do, and know of people in the industry with the same experience. SVM is an amazing algorithm that has been stymied by the slowness of most implementations, as well as being a memory hog. I’ll make a few posts in the future about it.
As an Engineering Recruiter, I learned early that Physics is pure engineering, but there were no JOBs and I was in Houston during the boom days. Actually one of my first placements who convinced a hiring manager that his Physics background and his signal telemetry instrumentation skills were transferable to Pipeline instrumentation...I got him the Interview. He did the rest and established a strong career in Houston
That couldn’t be more wrong. Actual Physics is as far from pure engineering as they come. The problem is that so many people don’t understand what Science is all about.
Nice start to XBT blog!
Thanks Bojan for highlighting the fact that powerful methods don't get "outdated" when you have all the shiny and parameter-heavy models around. Perhaps we should wait for your next posts, because this one did not help clarify where you've made success using XGBoost. Looking forward and kudos!
Great post!
My experience matches almost everything you have said...with the slight exception of "even specialized methods like support vector machines can be a better fit". I don't think I've had a situation where an SVM was the best fit in almost 15 years - do you ever still find it to be useful? Maybe I just haven't been in that type of specialized situation. (I still use Random Forests / GLM / GAMs, although not as often as XGB)
Yes, I indeed do, and know of people in the industry with the same experience. SVM is an amazing algorithm that has been stymied by the slowness of most implementations, as well as being a memory hog. I’ll make a few posts in the future about it.
Looking forward to reading and learning!
Try again https://www.data-cowboys.com/blog/which-machine-learning-classifiers-are-best-for-small-datasets
As an Engineering Recruiter, I learned early that Physics is pure engineering, but there were no JOBs and I was in Houston during the boom days. Actually one of my first placements who convinced a hiring manager that his Physics background and his signal telemetry instrumentation skills were transferable to Pipeline instrumentation...I got him the Interview. He did the rest and established a strong career in Houston
That couldn’t be more wrong. Actual Physics is as far from pure engineering as they come. The problem is that so many people don’t understand what Science is all about.