Finishing Renthop, beginning QuoraDup and Fedal Deep Dive

01 May 2017 . category: report . Comments
#kaggle #data science #tennis

My weekly report from the 24th of April to the 30th of April – an eventful week, finishing competitions and taking up new ones! Landed a good Top 15% position in the Renthop challenge, proud of it :)

Finally, the long two-and-a-half months of New York rental inquiries came to an end, and what a finish it was! Loved the ideas and feature engineering techniques of the top kagglers, they inspire me! I documented my ideation and execution process into an explainer blog, which did pretty well – Got published on a Medium publication and positive responses from Reddit and Twitter.

At work, i did some exploratory analysis on the Customer Health Score(CHS), identifying which patterns work and which don’t. Read up on “Hidden Markov Models” for the implementation of a better CHS system – the hidden states are “Green”, “Red” and “Yellow”, while the visible states are the activity on the platform. Sounded like a good idea for a couple of days, but after some playing around it doesn’t look like worth the effort. Will give it a try though :)

I’ve taken up the Quora Duplicate Questions challenge on Kaggle – One month left, this seems like a good time to push the accelerator! I put in a starter xgboost submission, but there is a lot more scope of feature engineering. Will explore into word vectors(Word2Vec) and apply Deep NLP concepts I’d learnt earlier! An exciting month of NLP ahead :)

Put on the backburner for a while now, the Fedal Deep Dive is back in action. I’ve decided to go ahead with just the Australian Open Final for now, the data is a little messy to be merged and cleaned. Currently, i’m through with some of the basic concepts(probably halfway) – Aiming for putting it up by end of week!

On the fitness scene, i’m back with the NTC workouts – the thighs and hips are feeling it :( Went for a couple of 1k runs, recovery on track :)

That’s it folks!

---