Kaggle 比赛技巧（四）
基本数据泄漏 Basic Data Leaks
Data leak
1.How bad it is ? 2.Whats the public opinions? 3.To exploit or not to exploit
Contents
1.Leakage types and examples 2.Competition specific. Leaderboard problem 3.Concrete walkthroughs
Leaks in time series
1.Split should be done on time a).In real life we don’t have information from future b).In competitions first thing to look: train/public/private split, is it on time? 2.Even when split by time, features may contain information about future. a).User history in CTR tasks b).Weather
Unexpected information
1.Meta data 2.Information in IDS 3.Row order
competitions:Dato, TalkingData, CAT
排行榜探测与不易察觉的数据泄漏 Leaderboard probing and Examples of Real Data Leaks
1.Types of LB probing 2.Categories tightly connected with ‘id’ are vulnerable to LB probing a).Company of user in RedHat competition b).Year, Month, Week in WestNile competition C:Truly Native 1.Predict whether the content in an HTML file is sponsored or not 2.Data leak in archive dates. But is it all? a).Data collection b).Date proxies C:Expedia 1.Predict hotel group a user is going to book 2.Data leak in distance feature 3.Reverse engineering true coordinates C:Pairwise tasks 1.Data leakage in item frequencies 2.Similarities from connectivity matrix