Did you know that you can navigate the posts by swiping left and right?
Model tuning and post-processing in RTO | Weekly Report 63
13 Nov 2017
. category:
report
.
Comments
#data science
#myntra
#kaggle
#fitness
#vlog
#tennis
My weekly report from the 6th of November to the 12th of November – a solid productive workweek and a little rollercoaster of a weekend.
I’m at that stage in my RTO project where my emotions, my moods are highly correlated with my model’s performance. Putting in last-ditch efforts to get to workable metrics! i added a couple more models with the hope of an outperforming ensemble, not yet fruitful. i read up on the concept of pseudolabelling and tried it on our dataset – it doesn’t quite work on supervised problems as well as it does on semi-supervised ones(less training data and a lot of unlabelled data). Feature engg is anyways a constant effort, we’re trying to bring down our false positives – trying to engineer the features on those lines. If all else fails, we might have a post-processing layer with human logic to minimize misclassification.
This problem is a troublemaker for all e-commerce companies alike, because of the high degree of randomness involved! It has been a real challenge to capture whatever systemic issues we can extract from our data.
Slight progress on the fast.ai course, though i got halted by some AWS issues. Confident of making good progress this week. I realized there’s no need to be working on a major kaggle competition alongside my work(as long as i’m learning!). Rather devoting myself to one real-world problem might give better dividends. However, i should be active on kaggle kernels and discussions no matter what!
A little confession – for the first time, i was counting off days to the weekend. Characteristically not me. Living for Friday nights? Nope. Fix it.
This week, i put out a run vlog – proud of a couple transitions in there, was a relatively easy edit.
On the fitness front, going strong! Put in the three runs, couple workouts and a tennis session. Less than a couple weeks to KTM, i’m gonna go on a longish long run this week. Plans to start squash too, excited!
That’s it folks. Later!
Shubhankar is an awesome person. He's Co-Founder at Houseware, building the next generation of Analytics. In his spare time, he likes to go out on runs!