1 of 7

Data Science Competitions: A know-how to participate

Ashis Kumer Biswas, Ph.D.

Assistant Professor, CSE @ CU Denver

It’s not the destination, it’s the journey.

2 of 7

Competitive Programming

Given a problem, solve it breaking the performance records than fellow competitors before a target deadline.

A Problem

Solve it

Breaking records

Before Deadline

Competition Venues

Tips to be overall competitive

Start early : read, read and read and discuss with partner(s). Understand the problem.
Quickly identify the problem type and solve using the dumbest idea you can first think of the moment.
Discipline yourself. It matters how many hours you put into it.
Do algorithm analysis. Choose a different algorithm, or customize an existing algorithm, a more advanced algorithm if you can think of anything better later. Go for incremental (tiny) improvements.
Master Programming language(s) and associated libraries.
Master the art of testing code.
Team work

Some Specific tips

Be proficient at preprocessing the dataset the best way possible including feature extractions, etc.
Learn to explore the given dataset. If possible gather some relevant domain knowledge if the field of problem is different than yours.
Learn to know about hyper-parameter tuning.
Learn about Ensemble strategies.
Learn to beat overfitting. Start develop good habits for each contest.
Research exhaustively! A winning submission constitutes many tiny, meticulously performed steps which otherwise would be ignored in a less competitive venue.
Practice, practice and practice!

Final tips

Set incremental goals
Watch/Review interviews of the winning entries.
Review most voted kernels/discussion notes.
Don’t worry about who are competing: are they PhD researchers, industry experts, etc?
Don’t worry about low rank. Focus and be determined to solve/submit your entry into the contest.
Participate in many competitions to develop your core skills.