@Sycorax: Depends a bit what you take the scope of data science to be.

The explosive growth of R has downsides too. What's the number of contributed packages? A language with 10 extras can't be taken very seriously as a general environment. One with 100 or 1000 or 10000 is more serious. But 100,000 would just be confusing to almost everyone. Also, R seems to be fracturing: already quite different ways to do graphics, and then there's tidyverse that seems to be polarising. github.com/matloff/TidyverseSkeptic is one sample.

Interesting link. @NickCox

I am still accepting advice on books, languages, libraries, career applications, .. and appreciate them

@Scortchi-ReinstateMonica what are the possible scopes of ds? How popular is each of them?

@NickCox what packages in R do you consider useful?

@nbro I never remember there is an AI.stackexchange site. Looks exciting

@Tim Join us there!

What popular applications do machine learning or AI have in industrial careers? Is Computer Vision similarly popular as NLP?

@nbro I hope so

I only knew datascience, now AI

You're asking two questions, which are quite vague. I don't think that NLP is more popular than CV, which is definitely quite important in AI

@Tim DS is just a buzz word. DS is basically statistics and machine learning performed by guys that do not really have any formal education in statistics

How NLP not more popular than CV? Are they both unpopular?

@Tim It's hard to answer this question, unless you perform some kind of survey

WHat languages and libraries are you using for AI or machine learning or statistics or datascience?

Right now, I am using mainly Python, but also a little bit of Matlab

what libraries in Python

Keras, TF, NumPy, etc

Do you recommend some books on doing AI/ML/Stat/DS in Python?

Honestly, I have never read such a book. I already knew Python before diving into AI or ML. I am sure there are useful books around though

I know Python too.

I'd like to know good pragmatic books on doing AI/ML/Stat/DS in Python

using popular libraries

@NickCox Yup. I personally find tidyverse not helpful or intuitive. Yes, once you know how to read it, it makes for easier pipelines, but first you have to learn it. R has a steep learning curve as it is, tidyverse just adds to it. It's an improvement for power users and an additional hurdle for novices, and I fear that people who rely heavily on it make R even more inaccessible for new users. Same, incidentally, for ggplot2 vs. base graphics.

@StephanKolassa I heard R for Data Science by Wickham is highly recommended. So I got a copy yesterday. How do you like it? books.google.com/…

What are base graphics? Are they in "base" package?

@Tim I have never read an R book, sorry. "Base graphics" are what you get with the plot(), lines(), points() etc. commands. You typically learn these first in R. (Unless you learn from a book written by a TBitGoG, a True Believer in the Grammar of Graphics. Then you will learn the GoG, a truly elegant way of plotting, and the first time you see base graphics somewhere, you become hopelessly confused.)

@Tim: for Python, I have skimmed "Hands-On Machine Learning with Scikit-Learn & TensorFlow" by Géron, which looked pretty encompassing.

Thanks. I will take a look

Reminds me of a joke. The learning curve is steep: if you put in a little effort, you learn a lot. Learning has to be on the y axis, naturally. I have to say that a lot of that data science book was underwhelming: have observations in rows and variables in columns? that's precisely the data structure I learned in the early 1970s, and it was banal even then

@Tim I hardly ever use R; I just opinionate on it occasionally. There's certainly an alternative universe in which I would use R a lot (ditto MATLAB, Python); but not one in which I would use SPSS a lot; that would be against the laws of physics. In this universe I use Stata almost exclusively, which doesn't mean that I don't respect much else. I skim lots of R books, but (oddly or not) mostly to see what I should be implementing in Stata, not to learn R.

I want to live in your universe too, but stata isn't much employable

for me not for you. Or I am just low information

So I have to settle with Python and R

I reviewed the 2/e of The Grammar of Graphics in jstatsoft.org/article/view/v017b03 -- and failed completely to foresee Hadley Wickham and ggplot2. Incidentally, Lee Wilkinson got in touch directly some years later regretting the tone of his reply to me. He's a very nice guy.

So you like ggplot2?

Again, let me stress: I don't use R routinely and I have never tried to use ggplot2. I read about it and look at examples produced with it. I can evaluate it selfishly in terms of what have I learned from ggplot2 that wasn't evident before from earlier work by Tukey, Cleveland, Tufte: almost nothing, except that occasionally I play with white grids on darker backdrops.

But any complicated syntax that isn't the complicated syntax you know is less than appealing. Stata's syntax for graphics no doubt looks utterly weird to any ggplot2 expert.

@NickCox In my book, steep learning curves are arduous precisely because learning is on the horizontal axis and effort on the vertical axis. (I always found economists' convention of putting price on the vertical and quantity on the horizontal axis confusing.) Then again, one could surmise that the English topography is not conducive to an intuitive understanding of steepness... (SCNR ;-)

@Stephan Kolassa I have publications on mountain features in Britain, I would have you know, including slopes that are nearly vertical.

@NickCox: "Mountain features" cough "in Britain". Very good. Reminds me of the Swiss high seas fleet. Britannia rule the peaks!

Sure, the Swiss win on mountains.

Although this isn't strictly related to this site, I think the following question needs more attention.