GDP => MINLP Transformation #3082
Conversation
There was a problem hiding this comment.
@michaelbynum you are reading our minds! We were just discussing about this transformation with @emma58 @jsiirola and @ZedongPeng
The code looks great, but since we have the chance of creating a new reformulation, should we use it to give it a good name? GDP to MINLP is not specific enough (as BM and Hull might do that as well). How about binary multiplication? Short and descriptive!
|
Great suggestion! I like it! |
Codecov ReportAttention:
Additional details and impacted files@@ Coverage Diff @@
## main #3082 +/- ##
==========================================
+ Coverage 88.28% 88.31% +0.03%
==========================================
Files 832 833 +1
Lines 92307 94881 +2574
==========================================
+ Hits 81495 83798 +2303
- Misses 10812 11083 +271
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. |
There was a problem hiding this comment.
Thanks so much for doing this, @michaelbynum! A few comments (most of which are my mistakes propagating, sorry...), but this generally looks good.
| rhs = 1 if parent_disjunct is None else parent_disjunct.binary_indicator_var | ||
| if obj.xor: | ||
| xorConstraint[index] = or_expr == rhs |
There was a problem hiding this comment.
Unfortunately (and this is my fault too), this isn't actually correct... It assumes that nested indicator variables are local, which is not necessarily true. The safest thing for now is to always set the rhs to 1 and to put the exactly-one constraint on the parent block (and transform from leaf to root, which I think you're already doing).
There was a problem hiding this comment.
Your point with putting the exactly-one constraint on the parent block (instead of the root disjunct's parent block) is that the exactly-one constraint needs to be transformed again if it is for a nested disjunction?
There was a problem hiding this comment.
Maybe I'm misunderstanding what the "root disjunct" is...
There was a problem hiding this comment.
Is putting the exactly-one constraint on the parent disjunct equivalent?
There was a problem hiding this comment.
Correct. It should be on the parent disjunct so that it does get transformed again.
|
Thanks for all the feedback, @emma58. |
Summary/Motivation:
This PR adds a GDP transformation to convert a GDP to a MINLP by converting f(x) <= 0 on a disjunct to f(x)*y <= 0.
Legal Acknowledgement
By contributing to this software project, I have read the contribution guide and agree to the following terms and conditions for my contribution: