mjachi
commented
2 weeks ago We've made the following the changes now.
- For Line 14, we added "influencing the global variance of the wind field"
- Line 25 has been replaced with "implementing an efficient method for synthetic turbulence generation based on a domain decomposition technique."
- Line 39-40 has been rephrased to "
DRDMannTurbis written in Python and leverages computationally powerful backend packages likenumpyandPyTorch." - Line 40 has been made more general: "The implementation makes DRD models easily portable to GPU and other backends via
PyTorch." - Line 42 has been changed to "but for which the source code may not be public or freely available"
In no particular order (as of https://github.com/openjournals/joss-reviews/issues/6838#issuecomment-2295424005)...
Line 14
For those not in the field this is not immediately physically relevant, consider adding few words here [fn1]?
Line 25
Maybe a few words describing this as well (I know its covered quite nicely a few lines later though).
Lines 39:40
It should be noted that these packages themselves utilize C / specialized frameworks for efficiently working on GPUs, otherwise it is a bit misleading, e.g. a pure PyPy package is also completely written in Python, without C altogether, which is not true of DRDMannTurb (or most other packages of course).
Line 40
This seems overly restrictive, Pytorch can provide connections to other backends as well, mentioning cuda seems a bit disingenous.
Line 42
I haven't read the papers, but if they say anything like "will provide code on reasonable request" then this needs to be rephrased (though I agree / understand the sentiment this comes from).
General comments
I am afraid for me, neural networks, and state of the art domain decomp without any other quantifiers / references feels rather overly general and weak, could these parts be elaborated on a bit?
Otherwise, it is definitely a nice piece of work :)
[fn1]: I liked https://spie.org/samples/FG02.pdf as a quick refresher...