This is just me here venting the frustration of participating in Kaggle competitions. First of all, Kaggle is an online data science competition platform where you are free to join any competition of your preference - tabular, imaging, NLP, time-series, reinforcement learning, etc. I have been participating at there for a while. I got some medals, but I just can't do a solo gold, when all my previous team members are progressing quickly. I have a family which constantly occupies me, and I feel like I am getting old and I can't think as sharp and quick as 10 years ago. To me, what Kagglers do really well is that they are "ALWAYS" able to observe the key issues of the problem and attack it with some simple, explainable, institutive tricks among other things. Unfortunately, people are selfish and even they may (mostly they don't) summarize their approach in a seemingly nice solution write-up, devils are in details and the top solution is often not reproducible. I don't blame them for keeping the tricks, which are their hard work and experience accumulated over many years if not decades. I just feel frustrated that no matter how hard I try, I still can't get to the top of the leaderboard. When others can always train a strong model to get to the top of the leaderboard, I am still blindly blending other people's public submission to make risky prediction. I feel meaningless for the blind approach, but I also can't train a strong model. I know a lot of people don't think it's worth the time and effort on Kaggle. For me, it's rather a training/thinking exercise and to compete with the best of the best in the world. It's an opportunity for me to prove myself just like people on TopCoder or Leetcode. If you are a fellow Kaggler (especially for competition master/grandmaster) and you get to read this post, please let me know. I would love to keep in touch with you. I don't even mind paying for your mentorship if you are willing to take me onboard. Thank you for reading this and listening. #data #datascience #kaggle
Kaggle is NOT data science. It is a modeling competition that is GAMEABLE. The same models and variants win over and over. It is not the same as competing in Leetcode or TopCoder. Edit: Also, I understand your frustration. Those who win are not winning bc they are better modelers than you. They are taking advantage of the way models are evaluated. Their models will never be productionized in a real world environment.
My point is that they are all online platforms for a certain skill sets. So I agree with you.
I had started with the tabular data competitions that are going on recently. I noticed almost all the top people from the leaderboard were doing nothing but blending models from other peoples solutions. While I was trying hard to do feature engineering and other model variants. This was very frustrating since most were just there to be on the leaderboard and not to learn lol.
Bro good on you for sticking to something. But just ditch and do leetcode
Thanks for listening and support bro