Tuning my stacker, and planning for the morrows | Weekly Report 37

24 Apr 2017 . category: report . Comments
#kaggle #data science #vlog

My weekly report from the 17th of April to the 23rd of April - a good week in terms of the leaderboard situation on kaggle – Jumped 700 odd places :) Still fine-tuning the model though.

In my desperate attempt to increase diversity in my stacker, i tried 6-7 algorithms(SVM, KNN,, GBM, ExtraTrees, even NaiveBayes!), but none were as good as the gbm or the xgboost. I also got LightGBM working on my system, the new workhouse for a lot of kagglers(it acted strangely slow for me). Anyways, there was a major leak on the competition (the timestamp in the images folder), which has boosted everyone’s score by almost 0.01! i also utilized a public script for my ensemble, some looking around is allowed eh?!

At work, i’m starting a project on churn prediction for our clients, and there are a hell lot of variables to consider! Sounds like an exciting plan :) i’ve to also build a retention model for the existing customers, tying back to feature adoption etc.

I’ve planned a lot of stuff for the next couple of weeks, on the top priority lies a Fedal Deep Dive video, i.e. an explainer of Fedal stroke-by-stroke dataset. Could be incredibly insightful in the styles of both players. Also planning to do a research(journalism blog) on the biases of different media houses(based on type of consumption). A very interesting problem! i’m currently learning D3, done with the basics!, hope to integrate my R workflow with the D3 systems.

I spent the whole of Saturday, going through an insanely mind-bending ridiculous post on Elon Musk’s new company – Neuralink. The absolute long-term vision is nothing short of genius!

This weekend, i spent a morning skateboarding and recording for a video at Cubbon Park. The editing is almost through, have to iron out a few chinks. Played clay court tennis at a new venue in Marathahalli, not great though.

Wrote this blog post in a semi-sleepy state. Forgive my errors :)

That’s it folks.

---