30 Days of Kaggle

At the start of this year, I embarked on my journey to learn AI/ML with the 30 days Kaggle challenge organized by ML Kolkata and in this blog, I will share the lessons learned everyday, share resources, problems faced along with how I solved those.

Day 1: Rising to contribution tier

Today’s task was to join the dedicated WhatsApp group and introduce myself and so I did. Next I created an account in Kaggle. After that, I followed the mentioned notebook and did as per instructions.

When an account is created, the profile is set default to novice performance tier and there are steps mentioned to rise in tiers. To know more about how to rise in tiers, check this out.

One of the problem that I faced, I couldn’t find the mentioned Data tab and I was confused how to submit for the competition. First make sure the notebook is public and then click on submit prediction. Choose the notebook option and click on submit. Checkout the below screenshot for reference.

Day 2 - 7 : Grasping concepts in Python

Day 2 was about getting introduced to python and learning about variables, data types and operators. Moving on to day 3, it was about learning about functions, methods, getting help with functions and arguments. Day 4 was learning about Booleans and the operators that are needed to work with it. Day 5 was getting introduced to data structures like lists and tuples, various methods of list, list comprehension and loops. Day 6 was about learning strings, dictionaries and their different methods. Day 7 was about learning to work with external libraries. Everyday there was lessons that need to be read and exercises to be completed.

Day 8 - 11 : Make or break week

Day 8 to 11 was about learning the basics of Machine Learning that is the Decision Tree and Random Forests with concepts like Overfitting and Underfitting, using pandas for Data exploration and model validation.

Day 12 - 14 : Moving up a level

Day 12 to 14 was about learning much complex concepts like the pipelines, how to handle missiong values, categorical variables, data leakage, cross validation and XGBoost model.

Day 15 - 30: Final competition

The final part of this challenge was the competition where we have to apply what we have learned in the first 15 days. So the contest was to predict the popularity of the songs where ceratin data are missing but to keep in mind that the model should neither be underfit or overfit and the error should be as less as possible [tends to 0].

Final results

I was procrastinating at the last day of submission and at one point I prioritized my sleep more than the submission. I woke up in the evening, made some changes, run the cells and made the final submission. And I stood 5th. Well I wanted nothing more and nothing less…

Swags

Thanks to Kaggle team for sending some swags to the top 5 participants.

Resources

• Python course

• Intro to ML course

• Intermediate ML course

• All Kaggle courses

• Checkout my codes for help

Experience

The challenge was divided into two parts: learning and competition.

Day 2-7 was quite easy given that I already knew Python and solving them took just 5 mins for each exercise. Day 8-11 was the “make or break” period and I was happy to take the challenge which I passed. Tbh, it was very easy. Day 12-14 was tough and I made it somehow.

Day 15-30 was the contest but I started the whole challenge from day 15th so by the time I started with the contest I had only 5 days in hand. Back then I was also writing this blog when I wrote as mentioned in image.

This is how I was in top 5 with lesser submissions. First I used random forest model learnt from the courses and made my first submission to see with that error value where I stand in the leaderboard, then I used all the models learnt in courses and compared the error and used the model with the lowest error. Then I went to the basic models and compared their rmse, where I got the lowest error of all the models.

And with this I would like to thank Rishiraj Acharya for helping me kickstart ny AI/ML journey.

If you come to know about this challenge from my blog, the good news is you can complete the courses. The necessary links are provided in Resources section and I have shared the answers as well.

More contests coming soon!!!