,
8 tweets,
2 min read
Read on Twitter
What machine learning tools do Kaggle champions use? We ran a survey among teams that ranked in the *top 5* of a competition since 2016.
The first question asked about the *primary* framework they used.
Very happy to see confirmation that winning teams prefer Keras :)
The first question asked about the *primary* framework they used.
Very happy to see confirmation that winning teams prefer Keras :)
The second question asked about secondary frameworks -- usually teams win with an ensemble that involves many different ML frameworks. Here are *all* frameworks used.
Sklearn tops that ranking: everyone uses sklearn (although often as an auxiliary, for preprocessing or scoring).
Sklearn tops that ranking: everyone uses sklearn (although often as an auxiliary, for preprocessing or scoring).
For the record, we got 59 data points from 2018, 48 from 2017, and 13 from 2016 -- the data is very contemporary.
These interesting results raise a question: is this just personal preference? Is there more? I think we're looking at a clear pattern.
These interesting results raise a question: is this just personal preference? Is there more? I think we're looking at a clear pattern.
Winning competitions or getting papers published is rarely about being the smartest or having the best idea from the start. Mostly, it's a game of idea refinement.
Winners are those who tried more things than everyone else, who refined their initial ideas further than the others
Winners are those who tried more things than everyone else, who refined their initial ideas further than the others
Winners are those who went through *more iterations* of the "loop of progress" -- going from an idea, to its implementation, to actionable results. So the winning teams are simply those able to run through this loop *faster*.
And this is were Keras gives you an edge.
And this is were Keras gives you an edge.
We often talk about how following UX best practices for API design makes Keras more accessible and easier to use, and how this helps beginners.
But those who stand to benefit most from good UX *aren't* the beginners. It's actually the very best practitioners in the world.
But those who stand to benefit most from good UX *aren't* the beginners. It's actually the very best practitioners in the world.
Because good UX reduces the overhead (development overhead & cognitive overhead) to setting up new experiments. It means you will be able to iterate faster. You will be able to try more ideas.
And ultimately, that's how you win competitions or get papers published.
And ultimately, that's how you win competitions or get papers published.
So I don't think it's mere personal preference if Kaggle champions are overwhelmingly using Keras.
Using Keras means you're more likely to win, and inversely, those who practice the sort of fast experimentation strategy that sets them up to win are more likely to prefer Keras.
Using Keras means you're more likely to win, and inversely, those who practice the sort of fast experimentation strategy that sets them up to win are more likely to prefer Keras.
