Feat/highs speedups - #898
Open
sjpfenninger wants to merge 2 commits into
Open
Conversation
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
sjpfenninger
requested review from
brynpickering and
irm-codebase
as code owners
September 9, 2026 06:45
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #898 +/- ##
==========================================
- Coverage 97.52% 97.38% -0.14%
==========================================
Files 39 39
Lines 5045 5092 +47
Branches 658 664 +6
==========================================
+ Hits 4920 4959 +39
- Misses 65 70 +5
- Partials 60 63 +3
🚀 New features to boost your workflow:
|
irm-codebase
requested changes
Sep 9, 2026
irm-codebase
left a comment
irm-codebase
left a comment
Member
There was a problem hiding this comment.
Looks good. Had to do some brain churning to understand the method.
Overall it looks good. I trust our tests catch poor implementation thanks to our static result checkers.
Requests:
- Modify comments and docstrings to get rid of chaff
- Make sure test coverage is complete.
Comment on lines
+93
to
+95
| Mirrors the broadcasting done by `_apply_func`, so that components added to | ||
| HiGHS in batches end up in arrays with the same shape, dims and coords as an | ||
| element-wise apply would produce. |
Member
There was a problem hiding this comment.
Two things:
- Wouldn't other backends benefit from this approach, if it is more efficient?
If it can be demonstrated (e.g., viamemray, we should consider moving this to the parent class. - Let us avoid referencing other functions in comments (
_apply_func). Names change and this is bound to become obsolete.
Comment on lines
+115
to
+118
| def _scatter_objects( | ||
| template: xr.DataArray, mask: np.ndarray, objs: list | ||
| ) -> xr.DataArray: | ||
| """Place backend objects at the `mask` positions of an otherwise-NaN array. |
Member
There was a problem hiding this comment.
Another generic function that perhaps move upwards if other backends would benefit from it.
Comment on lines
+150
to
+158
| status = self._instance.addCols( | ||
| n_new, | ||
| np.zeros(n_new), | ||
| np.asarray(lb_vals[mask], dtype=float), | ||
| np.asarray(ub_vals[mask], dtype=float), | ||
| 0, | ||
| np.zeros(n_new, dtype=np.int32), | ||
| np.empty(0, dtype=np.int32), | ||
| np.empty(0), |
Member
There was a problem hiding this comment.
Trying to summarise this to see if I got the behaviour right.
- total new variables to add
- objective coefficient (0).
- lower bound
- upper bound
- coefficient in constraints (0)
- constraint coefficient rows indices (none)
- coefficient values (none).
Basically, this puts the variables in the model, does not initialise or set coefficients for them in the objective or the constraint matrix. Kind of 'non-parametric' variable initialisation.
Comment on lines
+146
to
+147
| # Variables are added in one batch: `addVariable` allocates arrays and queries | ||
| # the model on every call, which dominates build time on large models. |
Member
There was a problem hiding this comment.
Do not refer to code in comments (addVariable) to avoid confusion in the future.
It's fine to just state that this is more efficient in HiGHS, and state what this actually does (create non-parametric variables in the model).
| summed, mirroring `highs_linear_expression.unique_elements`. HiGHS rejects a | ||
| row that references the same column twice, so the merge is required, not an | ||
| optimisation. | ||
|
|
Member
There was a problem hiding this comment.
Please adapt AI over explained text into actual docstrings.
Suggested change
| Entries are sorted by (row, column) and duplicate columns within a row are summed (3x + 3x + 3y = 6x + y). | |
| Otherwise, HiGHS would reject a row that references the same column twice. |
Member
There was a problem hiding this comment.
Github suggestions are so, so broken nowadays :(
Comment on lines
+256
to
+257
| # E.g. a coefficient whose absolute value is below `small_matrix_value`, | ||
| # which HiGHS drops, flagging it with a warning status. |
| lengths = np.empty(n_new, dtype=np.int64) | ||
| idxs: list[list[int]] = [] | ||
| vals: list[list[float]] = [] | ||
| for idx, expr in enumerate(exprs): |
Member
There was a problem hiding this comment.
Another attempt of trying to summarise what this is doing.
This is the row equivalent of _add_variable above (which uses addCols).
- fetches the bonds from the expression and builds its compressed sparse row matrix equivalent (to fit HiGHS). This is a 'squashed' version of the matrix constructed by the model, built for higher efficiency. Note that if the problem is very 'dense' this will be less efficient than regular notation. In most applications sparse will win, though.
- starts: where each constraint 'begins'
- cols: which variables are referenced (
$3x_1 + 2x_3$ -> [1, 3]) - coefs: the coefficients (
$3x_1 + 2x_3$ -> [3, 2])
- get the current number of rows (constraints) already in the model to make sure we do not forget / overwrite them.
- Add the set of new constraints in one go via
addRows- n_new: number of constraints to add
- lower / upper: constraint bounds
- coefs.size: total number of coefficients being added.
- starts: where each constraint starts
- cols: variable index for the coefficient in each constraint
- coefs: value of each coefficient
- Check if all is well, blow up otherwise.
- Return the location of the new constraints in the backend so we can extract results later.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #
Summary of changes in this pull request
Reviewer checklist