@elduvelle I use Python for everything but data analysis: Serial communication for interacting with hardware, GUI development, command line tools etc.
R is still unrivalled in the stats section. Anything modelling, it’s got you covered with well-documented packages from generalised mixed models, Bayesian models, change-point models, additive models. It’s all there, and accompanies the papers in which those methods are introduced - it’s bleeding edge stats. Oh, and then tidyverse, including ggplot of course.
But there’s also another component: In many degrees R is taught in the intro to stats courses, making it the first programming language one encounters. I was taught R in my Sports Science Bsc, I teach it to ecology and conservation students. I since learned other languages, but R is like coming home☺️ What I miss is good GPU support for pytorch-like processing though…
@elduvelle I liked using Matlab for a lot of data processing, particularly n-dimensional data where n>=3 or the data was really big (hours of millisecond-samples). Then after processing I used R for easier statistical modeling and for ggplot2. But eventually I moved everything into Python, which is my main tool now, though I'm no longer in cogsci. I'm happy with SciPy and matplotlib!
@elduvelle I have been using #R for data analysis and especially inference in neuroscience, medicine, and a couple of other fields. Some reasons why I don't or can't use #python specifically:
There are no python packages for non-parametric frequency/density inference, that is, inference where you are not assuming any parametric dependence, and in fact not even a functional dependence (as deep networks or Gaussian processes do instead, for example). Monte Carlo sampling is very important for this. R offers many great packages for this, especially #nimble
Working with R I only need to keep one version of the software and packages: the latest. Python is a headache with all the versions one must keep.
I find python's syntax illogical and clumsy compared to R's – but obviously this is completely subjective.
@elduvelle
Im primarily a python programmer but do most of my data analysis in R - the tidyverse and ggplot are absolutely lovely tools built in a charmingly cursed language
@elduvelle For processing, Python wins because of all the specialized libraries (e.g., MNE for EEG data), but for statistical analysis, R will always reign supreme
@kinleyid so you use both?
Is it that there are stat things that you can do in R that you can’t do in Python? (Which ones?) or is it that some stat things are simpler / easier in R? (For example?)
@elduvelle Yes, both! Matlab too actually, but less often.
I guess the big advantage of R is that it was built for stats from the start. The fact that dataframes and formulas are basic data types in the language makes everything straightforward. If I hear of a new stats technique, I know there will be an R package for it that will probably use familiar syntax, whereas neither feels guaranteed in Python. What about you, what do you use?
@kinleyid I’ve been been using Python and Matlab, I’ve usually found that you I can do all the stats I need in either of these, maybe just once I had to use a Matlab function for a test that wasn’t implemented in Python.
I followed some lectures on R and found it really confusing… but it was a long time ago.
Would you have an exemple of test that is implemented in R but not in the other languages?
@elduvelle Yes, R can definitely be confusing. I only learned it because I was so determined after Matlab told me I'd need to pay to do a t test.
An example: R's OpenMx seems to be able to do multilevel structural equation modeling, whereas a quick google search didn't return any equivalent Python library.
Add comment