Objective Sensitivity #282

andrewrosemberg · 2025-02-22T17:41:28Z

Pull Request Summary

Title: Objective Sensitivity

Description: This pull request implements functionality for obtaining the objective sensitivity with respect to parameters and the equivalent for reverse mode in the DiffOpt.jl package.

Additions: 335 lines
Deletions: 35 lines
Changed Files: 7
Commits: 11
Comments: 1
Review Comments: 4
State: Open
Mergeable: Yes (unstable)

Usage Example

# Always a good practice to clear previously set sensitivities
DiffOpt.empty_input_sensitivities!(model)

MOI.set(model, DiffOpt.ForwardConstraintSet(), ParameterRef(p), Parameter(3.0))
MOI.set(model, DiffOpt.ForwardConstraintSet(), ParameterRef(p_c), Parameter(3.0))
DiffOpt.forward_differentiate!(model)

MOI.get(model, DiffOpt.ForwardObjectiveSensitivity())

In the backward mode, we can calculate the parameter perturbation with respect to the objective perturbation:

# Always a good practice to clear previously set sensitivities
DiffOpt.empty_input_sensitivities!(model)

MOI.set(
    model,
    DiffOpt.ReverseObjectiveSensitivity(),
    0.1,
)

DiffOpt.reverse_differentiate!(model)

MOI.get(model, DiffOpt.ReverseConstraintSet(), ParameterRef(p))

test/nlp_program.jl

codecov · 2025-02-22T21:18:35Z

Codecov Report

Attention: Patch coverage is 88.88889% with 8 lines in your changes missing coverage. Please review.

Project coverage is 89.18%. Comparing base (93f058d) to head (a15f6f6).

Files with missing lines	Patch %	Lines
src/NonLinearProgram/NonLinearProgram.jl	83.33%	6 Missing ⚠️
src/diff_opt.jl	85.71%	1 Missing ⚠️
src/moi_wrapper.jl	88.88%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #282      +/-   ##
==========================================
+ Coverage   89.08%   89.18%   +0.10%     
==========================================
  Files          15       15              
  Lines        1969     2006      +37     
==========================================
+ Hits         1754     1789      +35     
- Misses        215      217       +2

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

andrewrosemberg · 2025-02-27T18:49:07Z

src/diff_opt.jl

@@ -199,6 +215,21 @@ struct ForwardConstraintDual <: MOI.AbstractConstraintAttribute end

 MOI.is_set_by_optimize(::ForwardConstraintDual) = true


codecov thinks this line is untested, but it is used internally in the tests.

andrewrosemberg · 2025-02-27T18:49:23Z

src/diff_opt.jl

+"""
+struct ForwardObjectiveSensitivity <: MOI.AbstractModelAttribute end
+
+MOI.is_set_by_optimize(::ForwardObjectiveSensitivity) = true


dito codecov mistake

….jl into ar/dualparameter

frapac

this PR looks good to me, overall!

frapac · 2025-03-11T15:58:11Z

docs/src/usage.md

+
+```
+
+Using Lagrandian duality we could already calculate the objective sensitivity with respect to parameters that appear in the RHS of the constraints (e.g, `cons` in this case for parameter `p`).


typo: Lagrandian -> Lagrangian

I would also mention explicitly that the objective sensitivity w.r.t. a parameter change in the RHS of the constraints is given by the optimal multiplier

frapac · 2025-03-11T15:59:37Z

docs/src/usage.md

+
+Using Lagrandian duality we could already calculate the objective sensitivity with respect to parameters that appear in the RHS of the constraints (e.g, `cons` in this case for parameter `p`).
+
+On the other hand, if the parameter appears in the LHS of the constraints, we can calculate the objective sensitivity with respect to the parameter using: the sensitivities of the variables with respect to the parameter, \( \frac{\partial x}{\partial p} \), and the gradient of the objective with respect to the variables \( \frac{\partial f}{\partial x} \):


I would suggest to mention that this is a consequence of the chain-rule

frapac · 2025-03-11T16:02:53Z

docs/src/usage.md

+MOI.get(model, DiffOpt.ReverseConstraintSet(), ParameterRef(p))
+```
+
+It is important to note that the (reverse) parameter perturbation given an objective perturbation is somewhat equivalent to the perturbation with respect to solution (since one can be calculated from the other). Therefore, one cannot set both the objective sensitivity (`DiffOpt.ReverseObjectiveSensitivity`) and the solution sensitivity (e.g. `DiffOpt.ReverseVariablePrimal`) at the same time.


what happens if we set the reverse parameter perturbation both for the objective and for the solution? Does DiffOpt returns an error, or does it keep the reverse parameter set the latest?

Nice catch, it returns an error, I will add a note here

frapac · 2025-03-11T16:05:27Z

test/nlp_program.jl

+    MOI.set(model, DiffOpt.ReverseVariablePrimal(), x, 1.0)
+
+    # Compute derivatives
+    @test_throws ErrorException DiffOpt.reverse_differentiate!(model)


implement dual of parameter anywhere

608ca25

andrewrosemberg requested a review from joaquimg February 22, 2025 17:41

andrewrosemberg commented Feb 22, 2025

View reviewed changes

test/nlp_program.jl Outdated Show resolved Hide resolved

andrewrosemberg added 3 commits February 22, 2025 15:47

fix api and more tests

c96e2c0

format

e459641

update docs

f243eb8

andrewrosemberg changed the title ~~Implement dual of parameter anywhere~~ [WIP] Objective Sensitivity Feb 25, 2025

andrewrosemberg added 2 commits February 27, 2025 10:48

update for appropriate API

0d2ec33

update documemtation

6d2efcf

andrewrosemberg changed the title ~~[WIP] Objective Sensitivity~~ Objective Sensitivity Feb 27, 2025

andrewrosemberg added 4 commits February 27, 2025 11:01

format

f366f2d

rm legacy function

1e2ac7c

rm unecessary functions

d4abb62

read necessary functions

d0bd2c6

andrewrosemberg commented Feb 27, 2025

View reviewed changes

andrewrosemberg and others added 4 commits February 28, 2025 15:37

Update DiffOpt.jl

a108599

fix bug indexing

ad3d72c

Merge branch 'ar/dualparameter' of github.com:andrewrosemberg/DiffOpt…

3e09271

….jl into ar/dualparameter

format

a15f6f6

frapac reviewed Mar 11, 2025

View reviewed changes

klamike mentioned this pull request Mar 25, 2025

Learning the dual and automatic DLL completion LearningToOptimize/LearningToOptimize.jl#31

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Objective Sensitivity #282

Objective Sensitivity #282

andrewrosemberg commented Feb 22, 2025 •

edited

Loading

codecov bot commented Feb 22, 2025 •

edited

Loading

andrewrosemberg Feb 27, 2025

andrewrosemberg Feb 27, 2025

frapac left a comment

frapac Mar 11, 2025

frapac Mar 11, 2025

frapac Mar 11, 2025

frapac Mar 11, 2025

andrewrosemberg Mar 11, 2025

frapac Mar 11, 2025

		@@ -199,6 +215,21 @@ struct ForwardConstraintDual <: MOI.AbstractConstraintAttribute end

		MOI.is_set_by_optimize(::ForwardConstraintDual) = true


		```

		Using Lagrandian duality we could already calculate the objective sensitivity with respect to parameters that appear in the RHS of the constraints (e.g, `cons` in this case for parameter `p`).


		Using Lagrandian duality we could already calculate the objective sensitivity with respect to parameters that appear in the RHS of the constraints (e.g, `cons` in this case for parameter `p`).

		On the other hand, if the parameter appears in the LHS of the constraints, we can calculate the objective sensitivity with respect to the parameter using: the sensitivities of the variables with respect to the parameter, \( \frac{\partial x}{\partial p} \), and the gradient of the objective with respect to the variables \( \frac{\partial f}{\partial x} \):

Objective Sensitivity #282

Are you sure you want to change the base?

Objective Sensitivity #282

Conversation

andrewrosemberg commented Feb 22, 2025 • edited Loading

Pull Request Summary

Usage Example

codecov bot commented Feb 22, 2025 • edited Loading

Codecov Report

Choose a reason for hiding this comment

Choose a reason for hiding this comment

frapac left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

andrewrosemberg commented Feb 22, 2025 •

edited

Loading

codecov bot commented Feb 22, 2025 •

edited

Loading