First Experience of Hackathon and Kaggle competition
I got really excited to join my first hackathon and my first Kaggle competition as well, I am kind of lost at the beginning, I ask the organiser some dumb questions like’ should l bring my lunch box, etc..
I went to the 400 Gerger st, Microsoft centre and after saw most people already in a team to join this competition, I am a bit worried nobody going to join my team as I am a total beginner when Amit and Sean asked me if they can sit next to me, I was pretty happy. After talking to them I found out that I am not alone, Amit is working for flight centre and Sean is first-year dentist student at UQ, Jia Qiu is 2nd year PhD student from Qut, none of us has much background on it, let alone the machine learning. it turns out that actually most of the members claimed that they are a beginner for AI, even though some of them have a strong IT background.
There are some issues with the Microsoft VM set up at the start, while Krist is working on it, Lex gave us a quick introduction on the Kaggle competition and how to get started to build a first linear regression model.
I just played around in Kaggle website and found that they are many tutorials under the discussion panel. I feel a bit not so nervous now once I knew there is something I can do. After discussed with my team members, we decided to kick off the competition by just follow one guild and digest it. Sean found one good one using Lassa, we just went through it on the first day. after that, I did a bit of research and find that XGboost is pretty good in terms of accuracy. I went to the leather board and click the top 20’s name one by one, looking for their tutorial, and found this one(linkage). I tried out its code without the data munging part, as I felt that part is overwhelming for me. it turns out the score is very low and gave me the rank of 2000 out of 4000 submissions. I thought it is the parameters not tuned as we knew that XGboost has tons of parameters. I found this introduction very helpful.
After finished reading, I tried a few changes and found that I only made things worse. I decide to keep the modelling part intact and working on the data munging, and finally ached pretty good score, rank 400. Oh yellllll…
The 2nd day afternoon, I am a bit tired and decide to go slow. I realised that data visualization is very useful and start to learn Matplotlib and Seaborn, etc.
At end of the day, surprisingly, we found out the score out team achieved is actually the highest one and won the hackathon, which really made me feel so happy.
All in all, it is great fun to join the hackathon to learn new skills and meet new people. our teams decided to work on some other projects together in the future, which I am really looking forward to.