Reverse engineer filtered and projected design variables #2462
Conversation
|
For a completely binary target structure: The difference (target-mapping(target)) after mapping would be: With the mini-optimization, the difference (target-mapping(optimized x)) looks like: where mapping includes filter with radius=0.1 and projection with beta=8. |
python/adjoint/reverse_design.py
Outdated
| from autograd import grad | ||
|
|
||
|
|
||
| def reverse_design(target, processing, maxiter=50): |
There was a problem hiding this comment.
Would be nice to have docstrings and typehints.
|
Codecov ReportAll modified and coverable lines are covered by tests ✅
❗ Your organization needs to install the Codecov GitHub app to enable full functionality. Additional details and impacted files@@ Coverage Diff @@
## master #2462 +/- ##
==========================================
+ Coverage 73.82% 73.94% +0.11%
==========================================
Files 17 18 +1
Lines 5269 5293 +24
==========================================
+ Hits 3890 3914 +24
Misses 1379 1379
|
python/adjoint/reverse_design.py
Outdated
| @@ -0,0 +1,52 @@ | |||
| import nlopt | |||
There was a problem hiding this comment.
Maybe change this to a lazy import inside the function, so that NLopt is only required if you want to run this particular function.
|
Probably add a test. |
python/tests/test_adjoint_solver.py
Outdated
| def test_reverse_design(self): | ||
| """Verifies that the reverse_design finds better initialization for specific | ||
| design than that design itself as initialization.""" | ||
| print("*** TESTING reverse_design ***") |
There was a problem hiding this comment.
This is too loose of a test.
I would do a gold master test where you just verify that you reproduce the results of a previous run that you checked manually.
|
Should pass the new Maybe not needed since the filtering is specified by a user function. |
python/tests/test_adjoint_solver.py
Outdated
| return projected_field.flatten() | ||
|
|
||
| reverse_x = mpa.reverse_design(target, processing) | ||
| tol = 1e-5 |
There was a problem hiding this comment.
Is this value supposed to match the ftol = 1e-5 in reverse_design.py?
There was a problem hiding this comment.
Not really. This value is just a small number to ensure future versions will have equally satisfying performance. I changed it to a smaller value.
Sometimes, it is desirable to have some specific initial guess for the topological optimization. With the presence of filtering and projection function, we can do a mini-optimization to reverse engineer the design variables whose filtered projection matches the desired structure.