Implement generating alternatives using metaheuristics #13

marnoldus · 2024-05-24T14:04:04Z

Pull request details

Describe the changes made in this pull request

Implements finding alternative solutions using metaheuristics. Creates an alternative generating problem from a JuMP model that can be solved by an algorithm implemented in Metaheuristics.jl.

List of related issues or pull requests

Closes #2

Collaboration confirmation

As a contributor I confirm

The documentation is up to date with the changes introduced in this Pull Request (or NA)
Tests are passing
Lint is passing

marnoldus · 2024-05-28T09:41:26Z

@abelsiqueira as you can see this pull request fails on Julia 1.6. The reason for this is that the Metaheuristics package I use requires a minimum Julia version of 1.7. Using an earlier version of Metaheuristics.jl gives errors in the current code. What do you think is the best way to resolve this? Should we add a compatibility requirement in the project.toml that julia should be at least 1.7? Then we should also update the continuous integration.

abelsiqueira · 2024-05-28T10:53:26Z

Hi @marnoldus, there are two possibilities:

What you described, change 1.6 to 1.7 in all places it appears.
Use Package extensions to only load the relevant metaheurisitics functions if the Metaheuristics package is loaded (which should be impossible in 1.6 because it's broken) and/or check the version number and act accordingly at package load time.

Solution 1 is simpler and straightforward. I prefer it.
The drawback is that the other packages are requiring 1.6, so there is some incompatibility - expectations are broken.

Solution 2 involves too many moving parts and is hard to maintain. It requires manual intervention when the LTS version of Julia changes, and it will create the false expectation that the package works normally on 1.6, which is not the case.

For me it's a clear choice to drop 1.6, but I'm tagging @clizbe for approval.

clizbe · 2024-05-28T11:11:03Z

Solution 1 sounds better. It just means the rest of Tulipa requires Julia 1.6 or later, while MGA requires Julia 1.7 or later, right?

abelsiqueira · 2024-05-28T11:26:04Z

Yes

gnawin · 2024-06-05T14:25:54Z

Thanks for all the work! I added some comments while reading the code for the first time. Most of them are about clarification from an outsider point of view, Although I can understand most after some googling/testing, but it would be nice if there can be more comments so I don't have to jump between things 😄 .

Not sure if you're still around to follow this up @marnoldus but I tag you both so you are both informed @greg-neustroev.

And there is an issue #22 with my comments for the existing code.

gnawin

Here's my review.

gnawin · 2024-06-05T14:11:53Z

src/alternative-metaheuristics.jl

+        fixed_variables::Dict{MOI.VariableIndex, Float64}
+    )
+
+Convert a constraint from a MathOptInterface function into a julia function of x. Supports only ScalarAffineFunction and VariableIndex constraints.


Calculating the value (not sure about this term) of a constraint instead of transforming a constraint? e.g., this function would return 3 when x=[1,2] for x_1 + x_2 >= 1; according to l27, it only supports ScalarAffineFunction?

This calculates the value of a scalar affine function that represents a constraint; it is based on the data in MOI.ScalarAffineFunction. Maybe there is an easier way to do that, e.g., a functioned called something like "apply constraint"? I'll investigate

would VectorAffineFunction work faster? I'll investigate

gnawin · 2024-06-05T14:12:53Z

src/alternative-metaheuristics.jl

+  index_map::Dict{Int64, Int64},
+  fixed_variables::Dict{MOI.VariableIndex, Float64},
+)
+  original_objective = JuMP.objective_function(model)


Suggested change

original_objective = JuMP.objective_function(model)

original_objective_function = JuMP.objective_function(model)

Will be resolved in a separate issue

src/alternative-metaheuristics.jl

gnawin · 2024-06-05T14:13:47Z

src/alternative-metaheuristics.jl

+  solution_values = collect(Float64, values(solution))
+  # Compute objective value or original LP.
+  original_objective_value =
+    extract_objective(original_objective, solution_values, index_map, fixed_variables)


can you just use something like objective_value(model)? Not fully clear why this new function.

This can probably indeed be done in a simpler manner, I will investigate

gnawin · 2024-06-05T14:14:10Z

src/alternative-metaheuristics.jl

+  # Obtain all constraints from model (including variable bounds).
+  constraints = JuMP.all_constraints(model, include_variable_in_set_constraints = true)
+  constraint_functions = map(c -> MOI.get(model, MOI.ConstraintFunction(), c), constraints)
+  constraint_sets = map(c -> MOI.get(model, MOI.ConstraintSet(), c), constraints)


what's a constraint function, what's a constraint set? Better comment on their definitions, e.g., with an example

These are defined as such in MOI documentation. What MOI.ConstraintFunction and MOI.ConstraintSet are is still not clear to me. Maybe you can add a comment explaining this, @marnoldus

This link explains them best: https://jump.dev/MathOptInterface.jl/stable/manual/constraints/
For linear constraints the Function basically desribes if it is a variable bound or constraint involving multiple parameters and corresponds to the lhs of the constraint, and the Set describes whether it's LessThan, GreaterThan or EqualTo and holds the rhs constant.

gnawin · 2024-06-05T14:18:18Z

src/alternative-metaheuristics.jl

+          push!(gx, c_set.lower - x[index_map[c_fun.value]])
+        end
+      else
+        throw(ArgumentError("Only linear non-vector constraints are supported."))


no quadractic function?

If it is possible to do this on a higher level, maybe it can be abstracted away. I'll investigate if MOI has methods for doing this in a more automated way

gnawin · 2024-06-05T14:18:32Z

src/alternative-metaheuristics.jl

+      end
+    end
+
+    # hx should contain 0.0 if no equality constraints extract_constraint


what about gx?

we can try a test with only equality constraints and see if this fails for fx. @marnoldus tagging you in case you know the answer

That's a mistake on my side, gx should also be [0.0] if no constraints are there.

@marnoldus could you fix this in this PR?

I fixed it. Just @ me again if I am needed elsewhere, but I probably won't respond again until wednesday or thursday

gnawin · 2024-06-05T14:19:29Z

src/alternative-metaheuristics.jl

+  metric::Distances.SemiMetric,
+  fixed_variables::Dict{MOI.VariableIndex, Float64},
+)
+  # Create map between variable indices in JuMP model and indices in individuals of Metaheuristic problem.


is mapping really needed? I can see a point, but it complicates things a bit in the thought process, and I'm sure if it is really needed.

This is converting JuMP indices to MOI indices. Maybe we can use JuMP indices directly or maybe this is implemented in MOI ?

gnawin · 2024-06-05T14:19:49Z

src/alternative-metaheuristics.jl

+  solution = minimizer(result)
+  function f(x)
+    f_old, gx, hx = objective(x)
+    fx = [f_old[1] - Distances.evaluate(metric, solution, x)]


shouldn't this be + instead of - if i compare with l69 in NearOptimalAlternatives.jl/src/alternative-optimisation.jl at 2-mga-metaheuristics · TulipaEnergy/NearOptimalAlternatives.jl (github.com)? And do we need to say max somewhere?

This is because in NearOptimalAlternatives.jl/src/alternative-optimisation.jl we are maximizing, but in Metaheuristics.jl we are always minimizing. I do agree that this introduces an inconsistency, though. Maybe we should be minimizing there.

test/test-alternative-metaheuristics.jl

Co-authored-by: Ni Wang <[email protected]>

greg-neustroev · 2024-06-17T11:50:58Z

src/alternative-metaheuristics.jl

+    # Initialise set of inequality constraints.
+    gx = Vector{Float64}(undef, 0)
+    # Add objective gap constraint depending on whether original LP is maximised or minimised.
+    if JuMP.objective_sense(model) == JuMP.MAX_SENSE


Can we pull this into a separate function, so that it can be reused in this file and in alternative-optimization.jl? @marnoldus ?

Adding constraints works very different for Metaheuristics (adding to array gx) or JuMP (@constraint), so I think it's not much cleaner when done in a separate function as the bodies of the if and else are very different

I suspected so, but the problem now is that if you want to change the formula, you need to change it in two places, although it is the same formula. We can think about how to unify these places in code in a separate issue.

src/alternative-metaheuristics.jl

gnawin · 2024-06-17T12:42:28Z

Great! Thanks for this PR.

marnoldus added 5 commits May 22, 2024 16:59

Implement normal flow metaheuristic alternative generation

b0d6a4e

Test and polish mga using metaheuristics

5e38332

Specify Metaheuristics version

f7c4eeb

Change Metaheuristics version

3f5ee41

Change Metaheuristics version again

e8073e8

marnoldus marked this pull request as draft May 28, 2024 11:55

marnoldus added 3 commits May 28, 2024 15:15

Resolve merge conflicts

e46b081

Resolve change from MGA to NearOptimalAlternatives

20598dd

Update Julia version to 1.7

ccbf958

marnoldus marked this pull request as ready for review May 28, 2024 13:41

marnoldus requested a review from greg-neustroev May 28, 2024 13:41

marnoldus mentioned this pull request May 29, 2024

Implement Particle Swarm Optimisation for Generating Alternatives #16

Draft

3 tasks

gnawin mentioned this pull request Jun 5, 2024

Make the names more understandable #22

Open

gnawin reviewed Jun 17, 2024

View reviewed changes

greg-neustroev and others added 3 commits June 17, 2024 13:14

Fix a typo

6e26bd5

Co-authored-by: Ni Wang <[email protected]>

Fix a typo

a2daa1d

Co-authored-by: Ni Wang <[email protected]>

Fix a description

4e2c16d

Co-authored-by: Ni Wang <[email protected]>

greg-neustroev reviewed Jun 17, 2024

View reviewed changes

This was referenced Jun 17, 2024

Resolve an inconsistency in how objective functions are defined #24

Open

Unify the objective function creation #27

Open

marnoldus commented Jun 17, 2024

View reviewed changes

src/alternative-metaheuristics.jl Outdated Show resolved Hide resolved

greg-neustroev mentioned this pull request Jun 17, 2024

Refactor constraint/objective function names #28

Open

marnoldus and others added 2 commits June 17, 2024 14:17

Set gx to [0.0] when no inequality constraints present

7ed1ba7

fix lint

8f9702b

gnawin merged commit 0c6eee8 into main Jun 17, 2024
9 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Implement generating alternatives using metaheuristics #13

Implement generating alternatives using metaheuristics #13

marnoldus commented May 24, 2024 •

edited

Loading

marnoldus commented May 28, 2024

abelsiqueira commented May 28, 2024

clizbe commented May 28, 2024

abelsiqueira commented May 28, 2024

gnawin commented Jun 5, 2024

gnawin left a comment

gnawin Jun 5, 2024

greg-neustroev Jun 17, 2024 •

edited

Loading

greg-neustroev Jun 17, 2024

gnawin Jun 5, 2024

greg-neustroev Jun 17, 2024

gnawin Jun 5, 2024

greg-neustroev Jun 17, 2024

gnawin Jun 5, 2024

greg-neustroev Jun 17, 2024

marnoldus Jun 17, 2024

gnawin Jun 5, 2024

greg-neustroev Jun 17, 2024

gnawin Jun 5, 2024

greg-neustroev Jun 17, 2024

marnoldus Jun 17, 2024

greg-neustroev Jun 17, 2024

marnoldus Jun 17, 2024

gnawin Jun 5, 2024

greg-neustroev Jun 17, 2024

gnawin Jun 5, 2024

greg-neustroev Jun 17, 2024

greg-neustroev Jun 17, 2024

marnoldus Jun 17, 2024

greg-neustroev Jun 17, 2024

gnawin commented Jun 17, 2024

	original_objective = JuMP.objective_function(model)
	original_objective_function = JuMP.objective_function(model)

Implement generating alternatives using metaheuristics #13

Implement generating alternatives using metaheuristics #13

Conversation

marnoldus commented May 24, 2024 • edited Loading

Pull request details

Describe the changes made in this pull request

List of related issues or pull requests

Collaboration confirmation

marnoldus commented May 28, 2024

abelsiqueira commented May 28, 2024

clizbe commented May 28, 2024

abelsiqueira commented May 28, 2024

gnawin commented Jun 5, 2024

gnawin left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

greg-neustroev Jun 17, 2024 • edited Loading

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

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

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

Choose a reason for hiding this comment

Choose a reason for hiding this comment

gnawin commented Jun 17, 2024

marnoldus commented May 24, 2024 •

edited

Loading

greg-neustroev Jun 17, 2024 •

edited

Loading