Finishing QuoraDup and beginning Indiahacks | Weekly Report 44

12 Jun 2017 . category: report . Comments
#kaggle #data science #hackerearth

My weekly report from the 5th of June to the 11th of June – a week of flux, a week where i took a rare break to recharge!

After a month of cracking my skull open at the Quora challenge on Kaggle, i’m free! Ranked a decent Top 12% amongst the 3400 participants, could’ve done better though. I put out a wrap-up blog on my approach towards the problem, a fairly long documentation of the thought process behind feature engineering and more! Although i wasn’t really tired, i took a not-so-well-deserved break for a couple of days. However that didn’t last long, i just couldn’t fend off the addiction for a new challenge!

Kaggle has got some serious challenges lined up currently, i’m going ahead with the Mercedes problem, given it is a regression problem and has a relatively short timeframe :) I played around with the dataset, read up on kernels/discussions but not much progress. Rather i’ve taken up the Indiahacks ML challenge for now, tackling a Market Segmentation problem for Hotstar and Road Sign prediction problem for Here. Three days of feature engineering on the Hotstar problem, and i’m within the top ranks. Competition on hackerearth is not the best, clearly :/

Elaborating a little on the Hotstar challenge, it’s basically a classification problem for a particular segment of users for whom the corresponding watch times, genres, aggregated time/day counts are given. Using a standard xgboost and a bit of creative feature engineering, i’ve got an AUC score of 0.8182. A majority of my important features involve the specific “show titles” watched and collective statistics related to it(What percentage of segment users watched “Ishaqzaade”, “What is the watch time of the India Pakistan match time in the segment vs non-segment users”).

The coming week is going to be my last at my internship at Belong, its been an amazing time! Currently wrapping up on my projects, and working side-by-side on the UserJourney hypothesis – well data cleaning, to be honest. We also finished up with the final PS-2 seminar recently, the last of my tests for college ;)

The Run plan is on-track and my knees are responding positively! Snuck in a couple workouts too, really bad at burpees though :( The tennis session was intense, so was Nadal’s La Decima. An incredible feat!

That’s it folks, later!

---