Moiv0.9 #41

jd-lara · 2019-08-23T21:49:50Z

This is pretty much work in progress to update the solver. See jump-dev/MathOptInterface.jl#736

jd-lara · 2019-08-23T21:57:34Z

@odow what's the new implementation of LQOI.VecLin?

    (LQOI.VecLin, MOI.Nonnegatives),
    (LQOI.VecLin, MOI.Nonpositives),
    (LQOI.VecLin, MOI.Zeros)

odow · 2019-08-23T22:00:14Z

They are removed. Use the full_bridge_optimizer instead.

jd-lara · 2019-08-23T22:02:03Z

Thanks, I haven't get there yet. The current changes are at Line 1361 of MOIWrapper

joaquimg

I will add another pass in a few hours!

src/xprs_attrs.jl

src/xprs_model.jl

src/MOIWrapper.jl

joaquimg · 2019-08-26T12:54:36Z

src/MOIWrapper.jl

+MOI.supports(::Optimizer, ::MOI.ObjectiveFunctionType) = true
+
+MOI.supports(::Optimizer, ::MOI.Name) = true
+MOI.supports(::Optimizer, ::MOI.Silent) = true


This does not work on Windows.
We can add a operational system check and send a warning explaning that the C API does not allow it. A callback must be implemented of the external file must be used.

needs testing

joaquimg · 2019-08-26T12:56:15Z

src/MOIWrapper.jl

+end
+
+function MOI.set(model::Optimizer, ::MOI.TimeLimitSec, limit::Real)
+    MOI.set(model, MOI.RawParameter("TimeLimit"), limit)


this won't work right?

needs testing

joaquimg · 2019-08-26T13:05:14Z

src/MOIWrapper.jl

+
+    return
+end
+=#


there is (TODO)

you can see docs of XPRSchgobj
the constant is the values associated with the index -1

to add the objective function: 25 + 5.3 * x_2 + 0.1 * x_5

mindex[0] = -1; mindex[1] = 2; mindex[2] = 5; obj[0] = 25.0; obj[1] = 5.3; obj[2] = 0.1; XPRSchgobj(prob,3,mindex,obj);

We can query this value with the OBJRHS attribute

Do you mean index 0?

https://www.fico.com/fico-xpress-optimization/docs/latest/solver/optimizer/HTML/XPRSchgobj.html

@joaquimg can you confirm here please?

joaquimg · 2019-08-26T13:08:29Z

src/MOIWrapper.jl

+        MOI.Semicontinuous{Float64}
+    else
+        nothing  # Also covers `NONE` and `LESS_THAN`.
+    end


do we need a specific value for existing_set ? cc @odow
we seem to be allowing type instability without necessity.

existing_set is used in the LowerBoundAlreadySet error. We could hard-code the throw into each branch, but it doesn't seem necessary. The compiler is probably smart enough. You could benchmark if interested.

src/MOIWrapper.jl

blegat · 2019-08-26T15:54:43Z

Call the file MOI_wrapper, not MOIWrapper

src/MOIWrapper.jl

Project.toml

src/MOIWrapper.jl

odow

You may want to look at the recent changes to Gurobi and GLPK.

Once I had written Gurobi, my approach to write GLPK was to comment everything out, and then slowly add a function at a time re-running the tests every couple of minutes. Then you can have something that consistently compiles, but just fails a heap of tests and slowly improve.

src/MOI_wrapper.jl

jd-lara · 2019-08-28T00:21:21Z

Made updates up to line 1718 and addressed most comments.

frapac

This review adresses minor typos in the new MOI wrapper of Xpress.

Manifest.toml

src/MOI_wrapper.jl

frapac · 2019-08-30T08:47:15Z

src/MOI_wrapper.jl

+end
+
+# ==============================================================================
+#    Callbacks in Gurobi


Gurobi -> Xpress?

Haven't got this far currently still at line 1718

frapac · 2019-08-30T08:50:04Z

src/MOI_wrapper.jl

+    throw(MOI.InvalidIndex(c))
+end
+
+function MOI.add_constraint(


Do you think it could be worthwile to add a bridge SOCToQuadBridge in MOI?

Not sure, Xpress supports both SOC and Quad and we could get to a point of supporting both

joaquimg · 2019-09-10T02:39:13Z

Any plans for updates here?

jd-lara · 2019-09-10T05:52:20Z

@joaquimg I made some progress, I pretty much ended up copying and pasting the code from the old wrapper for the Termination Status and found the equivalent in three different places for the raw strings. Please take a look, If I got the logic right from the old wrapper then the stop status is only used when specific termination status are triggered.

I am sort of stuck implementing the get_primal and get_dual functions starting in line 1922. My impression is that in Gurobi the values can be queried by individual variable and constraint (@odow?). However, in the old MOI wrapper (both in Gurobi and Xpress) the strategy was to return all of the primals into a container (get_linear_primal_solution!(m, x::Vector{Float64})) and Xpress doesn't seem to have the capability to get an attribute for a specific row and column as Gurobi has. Any pointers on how to proceed in this case? My first inclination would be to return all the solutions into a vector and then index into it. But, this will allocate a vector of all the solutions every time a specific variable or constraint gets queried.

Also looked into GLPK.jl and GLPK supports querying individual rows and constraints. Xpress seems to be limited to getting the solutions through getlpsol

jd-lara · 2019-09-10T06:16:54Z

The last line modified is 2020, this leaves the new wrapper pretty much at the same place as the old. The following sections are about problem modification and callbacks.

joaquimg · 2019-09-10T14:00:00Z

Lets get the entire solution vector from Xpress right after solve and then access it.
This is pretty reasonable and probably even more efficient.

odow · 2019-09-10T14:16:31Z

@odow should I copy the tests from Gurobi.jl as well and go from there?

Yes. The Gurobi tests, especially the MOI.Test ones are pretty comprehensive.

joaquimg

For the raw status I would map the actual xpress internal status to the string in the manual. And not the MOI statuses into detailed strings.

joaquimg · 2019-09-10T14:08:59Z

src/MOI_wrapper.jl

+    if length(f) != length(s)
+        error("Number of functions does not equal number of sets.")
+    end
+    canonicalized_functions = MOI.Utilities.canonical.(f)


Not sure if we need to cannonicalize. Might worth check in the manual

joaquimg · 2019-09-10T14:11:00Z

src/MOI_wrapper.jl

+            throw(MOI.ScalarFunctionConstantNotZero{Float64, eltype(f), eltype(s)}(fi.constant))
+        end
+        nnz += length(fi.terms)
+    end


Adding zeros might be faster than filtering

jd-lara · 2019-09-11T06:04:47Z

Lets get the entire solution vector from Xpress right after solve and then access it.
This is pretty reasonable and probably even more efficient.

My thinking about this is then to create a new field in model (previously names instance) struct with the complete solution and then query from there.

joaquimg · 2019-09-11T13:25:37Z

Vectors in the optimizer are ok. Struct with vector inside optimizer is also ok.

At some point I plan to only grab part of the solution. For instance, not querying slacks e not even duals, only primals. Then maybe do that in que lazy way. Just query the primal vector when the first primal is asked for. (None os this is needed now, just a heads up on future plans)

joaquimg · 2019-09-22T16:01:06Z

Let me know if you have more questions

jd-lara · 2019-09-22T19:23:05Z

I need to experiment more with the return objects of the get solution from Xpress. Also, the PSCC deadline is coming. I will get back to this after that.

joaquimg · 2019-09-27T13:11:53Z

Nice! I did imagine you could be distracted by PSCC, hope we can meet there next year ;)

jd-lara · 2019-10-09T19:06:36Z

@joaquimg is this needed in Xpress? I commented out for now to look for the Xpress equivalent of these settings.

    Note that we set the parameter `InfUnbdInfo` to `1` rather than the default
    of `0` so that we can query infeasibility certificates. Users are, however,
    free to over-ride this as follows `Optimizer(InfUndbInfo=0)`. In addition,
    we also set `QCPDual` to `1` to enable duals in QCPs. Users can override
    this by passing `Optimizer(QCPDual=0)`.

odow · 2019-10-09T19:45:38Z

is this needed in Xpress?

I think it's just a Gurobi specific thing

joaquimg · 2019-10-13T05:45:06Z

Exactly, that is a Gurobi thing.
The related xpress detail is that the presolve must be disabled if you want to query infeasibility certificates. I to that explicitly on MOI_Wrapper tests

joaquimg · 2019-10-13T05:50:48Z

src/xprs_env.jl

-            error("Invalid Xpress license")
-        else
-            error("Failed to create environment (error $ret).")
+mutable struct Env


not sure we should keep this...
it does not hold info like gurobi's, its just a boot in the library

jd-lara · 2019-10-28T21:56:52Z

closed in favor of #43

jd-lara added 4 commits August 22, 2019 17:31

add TOML files

f6e19dd

WIP: New MOI Wrapper

d74f669

WIP: Wrapper re-write

74b3b1e

update Manifest

8f30581

remove comment

638870a

blegat mentioned this pull request Aug 25, 2019

Update solvers to MOI v0.9 jump-dev/MathOptInterface.jl#736

Closed

20 tasks

joaquimg reviewed Aug 26, 2019

View reviewed changes

odow reviewed Aug 26, 2019

View reviewed changes

src/MOIWrapper.jl Show resolved Hide resolved

blegat reviewed Aug 26, 2019

View reviewed changes

src/MOIWrapper.jl Outdated Show resolved Hide resolved

blegat reviewed Aug 26, 2019

View reviewed changes

Project.toml Outdated Show resolved Hide resolved

joaquimg reviewed Aug 26, 2019

View reviewed changes

src/MOIWrapper.jl Outdated Show resolved Hide resolved

jd-lara added 9 commits August 27, 2019 14:48

file name change

47d8796

file reference change

ebfe37e

set MOI min compat version to 0.9.1

4db8a9f

add comment about the Gurobi in wrapper file

6f0bf53

create addnames function to add names in batch

78ae352

Define supported vector of variables sets

5ac113b

remove unnecesary method

d1a49f3

Address some PR comments

51b99b5

change OBJRHS syntax

5a58954

odow reviewed Aug 27, 2019

View reviewed changes

src/MOI_wrapper.jl Outdated Show resolved Hide resolved

src/MOI_wrapper.jl Outdated Show resolved Hide resolved

src/MOI_wrapper.jl Outdated Show resolved Hide resolved

src/MOI_wrapper.jl Outdated Show resolved Hide resolved

src/MOI_wrapper.jl Outdated Show resolved Hide resolved

jd-lara added 4 commits August 27, 2019 15:56

delete functions

3206a36

change implementation of variable type change

dfd9abe

update SingleVariable-in-Set constraint section

51123d9

add changes up to ### Optimize methods.

0ae31b7

frapac reviewed Aug 30, 2019

View reviewed changes

Termination status Changes

1cdb4e5

continue wrapper re-write

25085b1

address comments in the PR

26e3425

joaquimg reviewed Sep 10, 2019

View reviewed changes

jd-lara added 7 commits October 9, 2019 10:46

remove Compat

e1e011c

remove duplicate functions

cb8ec2f

Update tests

a8bd3a7

remove old require

433ddaf

update Environment definition

0186705

temporary test change

f73f3ca

update to wrapper

356e852

jd-lara added 2 commits October 9, 2019 12:14

add Test

d14aefe

enable first set of tests

b6eefe6

joaquimg reviewed Oct 13, 2019

View reviewed changes

jd-lara closed this Oct 28, 2019

Moiv0.9 #41

Moiv0.9 #41

Uh oh!

Conversation

jd-lara commented Aug 23, 2019

Uh oh!

jd-lara commented Aug 23, 2019

Uh oh!

odow commented Aug 23, 2019

Uh oh!

jd-lara commented Aug 23, 2019

Uh oh!

joaquimg left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

blegat commented Aug 26, 2019

Uh oh!

Uh oh!

Uh oh!

Uh oh!

odow left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jd-lara commented Aug 28, 2019

Uh oh!

frapac left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

joaquimg commented Sep 10, 2019

Uh oh!

jd-lara commented Sep 10, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jd-lara commented Sep 10, 2019 •

edited

Loading

jd-lara commented Sep 10, 2019 •

edited

Loading