Eugene Dubossarsky

1/ So here's my very handy and very simple method for creating a random subsample of a multivariate dataset that is very representative of the original. With added #machinelearning goodness ! Ready ? Here goes.

Agreed

St. Peter asks a newcomer: - Born? - Austro-Hungary. - Went to school? - Czechoslovakia - Married in? - Hungary - Kids born in? - The Third Reich - And your grandchildren? - USSR. - Where did you die? - Ukraine. - Man, you traveled a lot! - Nonsense, I never left Mukachevo.

Our next meetup on Tue, July 2 "Unlocking the Power of MLOps: Scalable and Sustainable Machine Learning" meetup.com/data-science-s…

Questions at academic conferences:

Our upcoming Late June event at Sydney University meetup.com/data-science-s…

Sydney AI+Data Science - our upcoming June event meetup.com/data-science-s…

I'll be honest I did not expect that the machines would start thinking and the humans would more or less just ignore the rather obvious fact that the machines are thinking

Data Science Sydney's November event is a deep dive into science, experimentation and A/B testing. Food and Drink provided as always, with plenty of time to network with practitioners and leaders. meetup.com/data-science-s…

Data Science Sydney is proud to host #kaggle grandmaster Xavier Conort ! Come for an exciting evening of feature engineering with Generative AI. Join me on the 27th of Sept meetu.ps/e/Mtljr/Wfl5v/i

We are proud to announce our seed funding round! Many 🙏to our awesome investors @MastryInc @collabfund @dwallerstein and brilliant advisors @Weather_West and @cargomoose This team is just getting started 👀 artemis.bm/news/reask-pla…

23.01.20 編集して再公開 Rで解析：GUI操作でggplot2のプロットが楽々作成「ggraptR」パッケージ karada-good.net/analyticsr/r-4…

@jim_savage_ That’s exactly what the unsupervised form of the random forest does. It distinguishes your data from multivariate uniform samples

The Random Forest Proximity Matrix and Neural Embedding both do essentially the same thing.

@WalterReade Very similar - but with a slightly different endpoint - in my case : many repeated splits / models, with the purpose of creating a very good split / AUC closest to 0.5.

@cargomoose Sounds the same as adversarial validation, unless I'm missing something. (But, it's a great thing to do!) fastml.com/adversarial-va…

1/ So here's my very handy and very simple method for creating a random subsample of a multivariate dataset that is very representative of the original. With added #machinelearning goodness ! Ready ? Here goes.

@peregrinari7 You don’t need to do that with RFs. Part of their unique magic is that you can use the entire dataset to train the model and then get Out Of Bag error on... the entire dataset again ! Kinda like k-fold cross-validation but baked into the alg

@MatterOfStats @teouchanalytics I would expect different error measures to introduce even more variety there

@teouchanalytics @cargomoose One thing it'd be interesting to see is the extent to which different ML algos selected different "best" samples (if, indeed, they do) and whether, across different sample sets, the same algos tended to agree or disagree. ie is "representativeness" well-defined. Fun times!

@ross_gayler :)