Darius Barušauskas 🇱🇹

4x kaggle GM, DL Researcher

[Darius Barušauskas 🇱🇹 @raddarGM]

@JFPuget Wow. finally 1st. huge congrats!

[Darius Barušauskas 🇱🇹 @raddarGM]

@abhi1thakur kaggle killer?

[Darius Barušauskas 🇱🇹 @raddarGM]

@tunguz you forgot to make a poll for that

[Darius Barušauskas 🇱🇹 @raddarGM]

@ShichaoLuo wow!

[Darius Barušauskas 🇱🇹 @raddarGM]

@mhajabri @tunguz second is cross-validation within cross-validation. You will figure it out :)

[Darius Barušauskas 🇱🇹 @raddarGM]

@mhajabri @tunguz Also there is two ways of doing it, you should try both. The less preferred method is with regularization. I first learned it here https://t.co/WO2YldvvDg (slide 26)

[Darius Barušauskas 🇱🇹 @raddarGM]

@mhajabri @tunguz There are a many examples on Kaggle where target encoding is done in a good and in a bad way. You just have to experience what data leakage is on your own. When you realize why it happens - it's very rewarding. I sure remember when I first learned do it the correct way.

[Darius Barušauskas 🇱🇹 @raddarGM]

@tunguz Would recommend to anyone to write it from scratch.

[Darius Barušauskas 🇱🇹 @raddarGM]

@tunguz Target encoding without data leakage.

[Darius Barušauskas 🇱🇹 @raddarGM]

@MeganRisdal @kaggle any spoilers?

[Darius Barušauskas 🇱🇹 @raddarGM]

@kaggle The position of it is very awkward though. I would have placed in near the leaderboard rank number. Extra 'Join' column looks like an overkill

[Darius Barušauskas 🇱🇹 @raddarGM]

@kaggle Wow. this feature was so long overdue! thanks

[Darius Barušauskas 🇱🇹 @raddarGM]

@tunguz which address should i mail my $2 to ?:D

[Darius Barušauskas 🇱🇹 @raddarGM]

@MeganRisdal I think gradio integration in any way would be very nice. Kaggle's ML experience ends at generating model files and predictions. I think that one final step is lacking - exposing models for use is a good experience for many.

[Darius Barušauskas 🇱🇹 @raddarGM]

@JFPuget Yes. I used SQL and R at that time. It was a dead end career. Went to do my own stuff with 100% python.

[Darius Barušauskas 🇱🇹 @raddarGM]

@sjodogne @CHULiege I remember when I discovered orthanc after using dcm4chee. It was like switching from windows 95 to windows 10 :)

[Darius Barušauskas 🇱🇹 @raddarGM]

@sjodogne @CHULiege Thank you for this amazing open-source contribution - you sure made a life of many companies easier!

[Darius Barušauskas 🇱🇹 @raddarGM]

@senkin13 it all looks good until you reach 0.800 mark. then it's a no man's land :D

[Darius Barušauskas 🇱🇹 @raddarGM]

@solidstate_birb @senkin13 not large. parquet files are less than 2gb