3 ms·

As an R user, my issue with Julia: what niche is it filling? R is established in statistics and data analysis, Python is established in general-purpose and data

As an R user, my issue with Julia: what niche is it filling? R is established in statistics and data analysis, Python is established in general-purpose and data analysis and ML, so why should someone learn Julia? The "main advantage" of Julia being faster than R seems irrelevant to me, R is plenty fast enough for my one-off statistical analyses. Further, I can't justify spending my attention on learning yet another language for analyzing data, given I already learned SAS, SPSS, R, STATA, Excel, and Python.

linspace 5y ago
People who is tired of wrapping C++ or Fortran libraries. It's not exactly a drop in replacement but it's almost there. If speed is not relevant to you then it may not be a good choice unless you are into language learning for the sake of it
bsdubernerd 5y ago
That's probably because you didn't try to write anything significant in R besides running an analysis. R is an interesting language which adapts "ok" to statistics and related fields, however it's very limited by today standards, completely REPL oriented and very, _very_ slow. In fact, the first thing one learns in R is to never loop on any set larger than a few hundred elements if you want your script to execute at decent speed. All the R speed is actually backed by fortran and C/C++ extensions. This goes from the R core to all the scientific and biosciences packages. R as a language has an extremely primitive interpreter and GC runtime. It's often much slower than python or perl. Although given the primitive GC and runtime, it's quite straightforward to write extensions for it. But *nobody* really wants to change language (whichever it is) to write extensions just to get some speed back. Especially when writing these extensions bars you from the entirety of the R ecosystem itself.
- WhompingWindows 5y ago
  Mmm, very good points. I used R for quite a few high-level statistical analyses, I found it handy for munging data and .rmds are nice for organizing a systematic set of chunks to make the code logical. Sometimes I ran into speed issues, which is when I would reach for data.table, which is quite a huge speed boost for some of R's slowest issues. I had to learn the basics of STATA for a modeling project, because the other researcher knew STATA (these network effects of collaborators knowing different things). Turns out fixed-effects multivariate logistic regression modeling is WAY faster in STATA than R, to the point where our infra just couldn't complete the R code in any reasonable timeframe so STATA was just the better pick. My main gripe is especially in healthcare, so many workers are trained in something in their 20's, then use that for decades. There are thousands out there still using SAS every day, for instance, because it's what they and their collaborators know. I suppose Julia and Python will grow in healthcare over time, SAS should in theory go down in usage...who knows!

linspace 5y ago

People who is tired of wrapping C++ or Fortran libraries. It's not exactly a drop in replacement but it's almost there. If speed is not relevant to you then it may not be a good choice unless you are into language learning for the sake of it

bsdubernerd 5y ago

That's probably because you didn't try to write anything significant in R besides running an analysis. R is an interesting language which adapts "ok" to statistics and related fields, however it's very limited by today standards, completely REPL oriented and very, _very_ slow. In fact, the first thing one learns in R is to never loop on any set larger than a few hundred elements if you want your script to execute at decent speed. All the R speed is actually backed by fortran and C/C++ extensions. This goes from the R core to all the scientific and biosciences packages. R as a language has an extremely primitive interpreter and GC runtime. It's often much slower than python or perl. Although given the primitive GC and runtime, it's quite straightforward to write extensions for it. But *nobody* really wants to change language (whichever it is) to write extensions just to get some speed back. Especially when writing these extensions bars you from the entirety of the R ecosystem itself.

WhompingWindows 5y ago
Mmm, very good points. I used R for quite a few high-level statistical analyses, I found it handy for munging data and .rmds are nice for organizing a systematic set of chunks to make the code logical. Sometimes I ran into speed issues, which is when I would reach for data.table, which is quite a huge speed boost for some of R's slowest issues. I had to learn the basics of STATA for a modeling project, because the other researcher knew STATA (these network effects of collaborators knowing different things). Turns out fixed-effects multivariate logistic regression modeling is WAY faster in STATA than R, to the point where our infra just couldn't complete the R code in any reasonable timeframe so STATA was just the better pick. My main gripe is especially in healthcare, so many workers are trained in something in their 20's, then use that for decades. There are thousands out there still using SAS every day, for instance, because it's what they and their collaborators know. I suppose Julia and Python will grow in healthcare over time, SAS should in theory go down in usage...who knows!