Summary of 2022 China Merchants fintech competition

This blog post is to record the feelings of the double non graduate students in this competition , And what we learned

Introduction to the game

FineTech The competition was held by China Merchants Bank , A total of three tracks have been developed 、 data 、 product , The competition rewards not only include bonuses , And internships offer, The time is the same every year ; I chose the familiar and unfamiliar data track race , I am familiar with it because I have participated in structural competitions before , Strange because last time I attended or 2021 It happened in 2000 , I am not born in the undergraduate class , I learned to climb electric poles when I was an undergraduate , What we are doing now is communication ;（ I took part in this competition , I went for internship and three-day tour , when , Even in 200 Within name , You may not receive an interview because of your background , But there are bonuses ）

The course of the game

The competition is divided into two stages a Bang and b A list of , The data is anonymous , The evaluation index is auc, Labels are listed 3:1：
a List features ： The line is consistent with the line , The distribution of test set and verification set is almost the same , So it's easy to get there online 0.9540 Score of , The difficulty is the construction of data features , Anonymous data has no specific representation , It is difficult to construct relevant features in combination with the data background （ But this competition , Some leaders shared their partial understanding of variable names , I have constructed two better features based on these cleavages ）.
b List features ： It all depends on guessing , The distribution of test set and verification set is inconsistent , Confrontation verification Of auc Can achieve 1, Finally, everyone is not structural features , But except Toxicant , Combine confrontation verification with distributed deletion Toxicant , Why is it called poison ？ Toxicant ： The distribution of training set and test set is similar , But if you keep it, you will lose points , such as ： against auc by 0.68, Offline 0.89, There are only 0.68; Some big guys try a feature , Offline 0.83, Online direct 0.86, Direct front 20; Frankly speaking , The big guys are still stable , Although it depends on luck , But reasonable screening can still get high scores .

feel

excited ——> Full of energy ——> Looking forward to the opportunity ——> disappointment