Closed nschloe closed 3 years ago
The domains names are too coarse. Perhaps they can be added as SVG in inkscape ("add text"). Also, svgo shrinks the image even more:
Performance.svg:
Done in 639 ms!
534.776 KiB - 49.7% = 268.985 Ki
(npm install svgo
.)
Okay, I'll increase the text resolution after I figure out how to get the other figure to save in full screen. The numbers are appearing all crunched underneath
got it
figure to save in full screen.
Don't bother with "full screen". You can scale the SVG after save. If you want more or less space between the tick labels, decrease/increase the font size. Remember you can always scale the svg afterwards.
I see. Yea, this is much better to be able to rescale as I please afterwards .
Okay both images are updated. I applied the optimization using svgo.
Alright one more fix to the benchmarks to annotate the clipped timings for the serial executions.
Okay, got to run now. Best finish up the figures, and it should be good to go. I won't be too strict anymore. I'll have whedon generate the PDF, have a last look and then approve. I will mention something about the good quality of the software and the article, and I'll probably also mention that I'm not the biggest fan of (seemingly) domain-specific mesh generators. Anyway, it'll pass tomorrow and you can enjoy some Christmas holidays.
Thank you Nico! I'll finish up the figures and merge those changes in. Merry Christmas and happy holidays to you as well!
Okay, added my final (positive) review. Congratulations! I also would like to tell you personally how well I liked your responsiveness. I asked you to do many more things than I would normally have because I felt my remarks fell on fertile soil. Hopefully you got something out of it, too.
Also, say hello to Antoine if you see him around! (We were office neighbors in Berlin a few years back.)
Thank you! I've learned a huge amount from your guidance in this process. The most productive review I've had to date.
Antoine and I work together actively on this project here in Brasil. He also mentioned you and says hello!
What's "high-geometric"? Perhaps you mean "high geometric quality", though I'm not sure what "geometric" wants to say. Is there a mesh quality other than geometric? (Genuine question.)
triangles/tetrahedral elements --> triangular/tetrahedral elements
Finite Element Method ---> finite element method, likewise Full Waveform Inversion etc.
"This in part contributes to the reality that automatic mesh generation for geophysical domains largely still remains an unsolved problem." The wording unsolved problem is too big. The Riemann hypothesis is an unsolved problem. The absence of a function that translates a data file to a density function is a missing tool. I'd rephrase this as (suggestion) "mesh generation for geo domains is not user-friendly".
"However, the usage of a particular sizing function"
Not the usage of a particular sizing function is an issue, but any sizing function that benefits from vectorization.
Let's just call this Python API or Python application programming interface.
How so? I thought in geo domains, you use cubes all the time.
"the circumcircle radius divided by the incircle radius"
It's the other way around.
I would like to have something like this: (not perfectly written) "In mesh generation, there is always a trade-off between generation speed and mesh quality. Since every generator draws the line elsewhere, or might even have user parameters to increase quality at the expense of time, it can be difficult to compare the results. That being said, with default settings Gmsh will produce high-quality meshes by far the fastest, SeismicMesh will produce meshes with the best quality, but much slower. Gmsh becomes comparatively slow when a user-defined mesh-density function is involved. This is exactly SeismicMesh's use case."
"Results indicate that Python array manipulations dominate parallel execution time and inhibit scalability"
I'd like to know if this is intrinsic to the problem or if it can be fixed or if it is unclear.