Translate empty row rule and reduction to apply! interface #13
Conversation
Codecov Report
@@ Coverage Diff @@
## master #13 +/- ##
==========================================
+ Coverage 46.56% 46.83% +0.27%
==========================================
Files 13 14 +1
Lines 902 901 -1
==========================================
+ Hits 420 422 +2
+ Misses 482 479 -3
Continue to review full report at Codecov.
|
…dn't make the port over from Tulip.
src/lp/empty_row.jl
Outdated
| Sets the dual variable $y\_lower_{i} = \max\{0,y\}$ for the greater-than | ||
| constraint and the dual variable $y\_upper_{i} = \max\{0,-y\}$ for the | ||
| less-than constraint. |
There was a problem hiding this comment.
Small clarification.
Unless row i is infeasible, both dual multipliers y_l and y_u are zero: this corresponds to the reduction EmptyRow(i, zero(T)), which is created in the non-infeasible case.
Thus, the y_l = max(0, y), y_u = max(0, -y) in the post-solve only yield non-zero values if the row was infeasible.
My rationale for initially doing so was to have "status-agnostic" reductions and postsolve.
There was a problem hiding this comment.
Updated the doc, let me know what you think.
src/lp/empty_row.jl
Outdated
| We eliminate row $i$ if, for each variable index $j$, | ||
| ``math | ||
| | a_{i,j} | ≤ ϵ | ||
| ``` | ||
| for prescribed tolerance $ϵ$. |
There was a problem hiding this comment.
I don't see this being done in the code, is it on purpose?
There was a problem hiding this comment.
Hmm no, I wrote this first and then never followed through on it. Do we want to do this? If so, at what point? It feels a bit weird to do it in load_problem!: if the user is passing us the matrix with small nonzeros, it seems like we should respect that.
There was a problem hiding this comment.
If we do it, I would rather put it in a different rule, with a drop tolerance for rounding small values to zero and large ones to infinity.
As for the where, once at the beginning of presolve seems reasonable. See also #12 (comment)
Co-authored-by: mtanneau <9593025+mtanneau@users.noreply.github.com>
…compute row sparsity
Co-authored-by: mtanneau <9593025+mtanneau@users.noreply.github.com>
No description provided.