Also I want to shamelessly plug my upcoming book. ...

# datascience

[thomasnield]

Also I want to shamelessly plug my upcoming book. I just finished writing the first draft but first 5 chapters are up on O'Reilly. https://www.oreilly.com/library/view/essential-math-for/9781098102920/

[Michal Harakal]

It would be interesting to see some Kotlin code in your book, anyway I am looking for the book

[thomasnield]

Thank you. I had to do this one in Python, although I did give merit to Kotlin in the final chapter. I think Kotlin's best bet is to position for the post-data science transition that's already happening. I talk about that extensively in the final chapter.

[Filipe Duarte]

Is this transition related to the shift from more research-oriented to engineering discipline?

[thomasnield]

@Filipe Duarte I think that's a large part of it, along with the realization most "data science" work is just shadow IT getting mainstream acceptance. https://veekaybee.github.io/2019/02/13/data-science-is-different/

[thomasnield]

I think "data science" largely eroded the centralization of IT, and that's probably the most practical impact it had outside of Silicon Valley. https://www.wsj.com/articles/get-rid-of-the-it-department-11637605133

[thomasnield]

The reality of "deep learning" and the AI renaissance is most companies are not equipped to do it, because manually labelled data is expensive especially if you're not Google or Microsoft. https://www.nytimes.com/2019/08/16/technology/ai-humans.html

[thomasnield]

I also think we are heading into another AI winter personally. https://www.thelancet.com/journals/landig/article/PIIS2589-7500(21)00208-9/fulltext

[thomasnield]

https://medium.com/starsky-robotics-blog/the-end-of-starsky-robotics-acb8a6a8a5f5

[thomasnield]

https://spectrum.ieee.org/andrew-ng-xrays-the-ai-hype

[thomasnield]

The number of data scientists I have met with little coding or software engineering skills is astounding, and I think this reason and others is why we will see less "data scientist" job titles in the coming years.

[Michal Harakal]

@thomasnield So what is your career recommendation for an aspiring data scientists coming a from software development ?

[thomasnield]

@Michal Harakal I think software engineers are in a much better place than most to enter a "data science" career. Get good at SQL and working with data on a larger scale. Focus more on statistical learning rather than machine learning. Be curious where data comes from and not just what it says. Also realize the best jobs are probably not called "data scientist" anymore, but rather have a role on something more specific.

[thomasnield]

Also have awareness of industry application. 90% accuracy recognizing faces on social media is a nice trick, but if it's a surveillance system or worse a military application, 90% accuracy is absolutely awful. I think this reluctance to realize that machine learning application requires a high tolerance for false positives/false negatives is something that needs more conversation.

[thomasnield]

https://www.nytimes.com/2020/06/24/technology/facial-recognition-arrest.html

[thomasnield]

https://www.newsweek.com/black-teen-barred-skating-rink-after-facial-recognition-camera-misidentified-her-1610023

[Filipe Duarte]

I agree with you. Some jobs offers are changing from “data scientist” to something more specific such as machine learning engineering, data engineering, risk data scientist, model & risk data science...

[Filipe Duarte]

With this career evolution, will Kotlin be more important? Because of Software engineering stuff? Language more robust to write software for data science and machine learning?

[thomasnield]

@Filipe Duarte I think Kotlin does have an opportunity because of the software engineering stuff. It takes time for industries to adopt a new platform though.

[altavir]

A lot of time. But for me ready visualization and surprisingly Jupyter (never believed in notebooks before) are game changers. We can do everything now in Kotlin without pythonic parts.

[Filipe Duarte]

Sure! What is the jupyter kernel recommended for Kotlin?

[altavir]

https://github.com/Kotlin/kotlin-jupyter

[thomasnield]

I need to play with that too. I've always avoided notebooks but I got to give them another chance.

[altavir]

Here is an example from today lecture (sorry, not a lot of comments): https://datalore.jetbrains.com/view/notebook/tXEHOkSoH0wVt9lLD0Nd5l. Brownian motion simulation plus Monte-Carlo integration.