Torch/Keras in R
the ecosystem moves forward
Recently, I more formally picked up machine learning again - in a more day-to-day fashion rather than one-off projects. Specifically, I’m trying to use R within the context of deep learning. Given that the dominant ecosystem for machine learning mostly remains python this might come as a surprise.
Thing is, I don’t like pivoting in and out of environments. In particular, data wrangling is much easier in R than in python. Given that a good amount of machine learning is dealing with cleaning data and data wrangling it often seems a better fit to use R rather than python (with the exception being image processing).
Anyway, I tried such a deep dive a couple of years ago with the excellent book “Deep Learning with R” by Francois Cholet. Sadly, at the time, the R ecosystem of Keras felt very brittle and incomplete. But, revisiting Keras, and Torch, again after a couple of years was a breath of fresh air. The R APIs feel more feature complete and in line with the python one.
The last couple of days I’ve really enjoyed the combination of {torch} and {luz} to model ecosystem flux data and phenology. Some key resources have been the Posit.io (rstudio) machine learning blog - which reblogs some of the work in the “Deep Learning and Scientific Computing with R torch” by Sigrid Keydana. Given that the first book by Francois Cholet was brilliant, I assume the second version is better with the increasing ability of TensorFlow and the Keras APIs. Pick up either of these books to get a good handle on machine learning in R. Depending on if and where I find time, I might blog about some tips and tricks.
Reuse
Citation
@online{hufkens2023,
author = {Hufkens, Koen},
title = {Torch/Keras in {R}},
date = {2023-10-30},
url = {https://khufkens.com/posts/machine-learning-torch-R-luz/},
langid = {en}
}