Christopher Akiki

4 million @lichess chess puzzles

@anishgiri @lichess

Please tag the AI company you want me to partner with, in comments.

Sad story in two acts.

@sarahookr We had bata in Lebanon too!

I grew up with bata shoes in eswatini and in Mozambique. Had no idea it was such a universal brand.

@MeganRisdal You absolutely do not owe me anything, but I would LOVE some Kaggle swag ❤️

@christopher PS have we sent you any Kaggle swag yet? We def owe you!

Do you want to host a hackathon on Kaggle? We're launching self-serve tooling soon. Let me know if you want early access!

@kylelostat What's it feel like?

fml, claude is back up but I’m out touching grass

@MeganRisdal Kaggle swag for a new boyfriend? 😁

My boyfriend broke my beloved NVIDIA coffee mug. I will trade Kaggle swag for a new one. 🥺

@MeganRisdal Just saw the ping on discord! Very cool!

🎮🃏🐺♟️

@julien_c Can i get ssh access to your claude code setup?

spent the last 3 days perfecting my claude code setup, ama

@MeganRisdal Thank you Meg! I'll start a post in the forums :)

@christopher But if you want to just post details of the issue with this file here I can also share with the team & get back to you on Twitter, too :)

TIL: @kaggle frontend now parses .parquet files and renders them!

UMAP connectivity plots of 3,627 chess openings from the @lichess datasets (hf.co/datasets/Liche…)

@JamesABednar Isn't there "more data" where more edges are bundled?

@christopher This data is monotonic without a central null value, so I'd argue against using any diverging colormap (coolwarm, managua) since they create false patterns in the data. Binary also looks misleading. For the rest, just a matter of preference!

Which colormap do you think looks the nicest? I'm leaning toward plasma.

@leland_mcinnes Oh interesting, so managua's hues but dark to light?

@christopher Plasma or viridis probably. There is something interesting about the managua though -- I think finishing at a blue hue might be more interesting than yellow. I wonder if you could make a sequential rather than diverging version that runs through a similar hue range?

Removing the single-author-ball lets the plot breathe. More structure in this version.

4M @dblp_org authors with 29M co-authorship edges.

@lauriewired @TheZachMueller Your new rabbit hole?

The “opposite” of a GPU is kinda weird. Graphcore has a strange chip (IPU) that loves tiny batches...but is also massively parallel. It used to cost ~$100k. Now you can get one on Ebay for $160 bucks. The catch is it's almost impossible to use.

@TheZachMueller @LambdaAPI Not even sure it's in scope for y'all, but beefy CPU-only instances for data-intensive workflows would be cool.

How could @LambdaAPI improve the platform for you? What QoL things can I help with in the new year? Let me know ⬇️