Weighted Volumes #406
Weighted Volumes #406
Conversation
Codecov Report
@@ Coverage Diff @@
## develop #406 +/- ##
===========================================
- Coverage 88.75% 88.73% -0.03%
===========================================
Files 121 121
Lines 10676 10762 +86
===========================================
+ Hits 9476 9550 +74
- Misses 1200 1212 +12
📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
|
Rebased with latest develop. After some consideration of the axis ordering I was able to get a reasonably positive comparison between the MATLAB workspace data and this changeset. Notebook attached. There are still some differences in high frequency, but I think that is minor differences in the 3D reconstruction itself, along with my very crude comparison, and not due to this feature/changeset. I am going to move forward with trying to refactor the class structure such that the single volume case is the subclass. |
|
stashed notebook in my fork ( |
|
@janden , I still want to check the scaling and add some notes regarding the references we spoke about, but if you want to take a cursory review of the draft while I'm away that might save some turnaround time when I return. I'm sure there are things to change... |
|
Updated notebook comparing the MATLAB Benchmark data and Python implementation can be found here: It is not included in this PR due to size. There is a fresh issue created to construct a gallery style tutorial once all the supporting code has melded into |
|
Rebased this with the latest |
|
Just refreshing this guy, a CI change took effect. |
|
Moving to v0.11.1. Should be a good time to work through this review since I will be stuck on basis/cov dev for a while. I will update the code again after the v0.11.0 release. |
|
Revived this at |
this branch was mostly just merge updates at this point.
string replacement and comments
|
@janden , thanks again for the review. Other than the |
|
Almost there! |
|
Yay! 🧟 |
Related to #405 .
Stubbing in the weighted volume code. Currently this is just extending the existing volume estimation codes with two sub-classes, mainly to shake out any issues or missing content bringing the feature over to Python.
There is either a tolerance or determinism issue I hope to narrow down on today. Other than that it is able to reproduce an array of
rvolumes given uniformn-by-rmatrix of1/sqrt(n)weights like we discussed last meeting.We should probably make sure the python code (including the unweighted codes..) actually does what it is supposed to do before I spend time on next steps/documenting. For example, we discussed potentially making the parent class generalized (weighted) and instead sub class the unweighted code as a special (child) case once things are working. Before we do that sort of work it might be good to reproduce some of the Matlab results...
This seems like a good place to sanity check.