adding alternative mode volume calculation for lower memory usage #287
Conversation
|
Spell check passed successfully for 1 notebook(s). |
There was a problem hiding this comment.
Thanks @FilipeFcp looks good to me overall. Only thing I'm wondering is whether the interval_space change was necessary? It seems like you changed from coarse spatial sampling to coarse time-domain sampling, with a net memory decrease of a factor of 2, which sounds reasonable but I'm wondering keeping the coarser spatial sampling would change the results significantly?
tomflexcompute
left a comment
tomflexcompute
left a comment
There was a problem hiding this comment.
That's great @FilipeFcp . Seems like there is a 10-20% difference in the calculated mode volume. Are there results in the literature where we can benchmark the results?
|
Thanks @yaugenst-flex It doesnt change significantly, but it brings the results closer to those reported in the paper. For this type of calculation, I believe the error introduced by using the raw data is preferable to the loss of data caused by downsampling. I added a comment regarding the interval_space in the notebook. |
|
@tomflexcompute Yes, the structure is from this paper. They report 0.39 (lda/n)^3 |
Great that's very good! |
|
@FilipeFcp if you could squash these commits it’s good to go |
Hi all.
Since a user used this notebook as a reference and had issues with kernel crashing during mode volume calculation, I made a small modification to reduce the memory usage, and also added a small discussion on how to use raw monitor data (without expanding symmetries) to calculate it.