Bridge backend of Model if needed #2610

odow · 2021-05-21T04:10:47Z

This PR addresses the "time-to-first-solve" issue (jump-dev/MathOptInterface.jl#1313). It is essentially a repeat of jump-dev/MathOptInterface.jl#1252, but at the JuMP level instead of MOI.

Closes

Closes #2520
Closes #1627

Background

The main issues are:

Bridges are good, and have a small run-time cost if un-used
Due to the design, bridges are not type stable. The main reasons are:
- New bridges can be added by the user
- Variable bridges can return a function of a variable, instead of just a variable
- Determining whether a variable is bridged requires a run-time lookup.
Because of the type instability, bridges don't precompile well, and they have a large compile-time cost.
This compile time-cost happens even if the bridges are not used. As one example:
- To set the objective, we must first delete the current objective. Because we can only tell at runtime if the current objective is bridged, inference starts exploring the branch that deals with the bridges. Objective bridges lead to the graph, which leads to other bridges, and so inference continues until we hit the type-instabilities and it gives up.
One attempt at solving this was Add auto_bridge to CachingOptimizer MathOptInterface.jl#1252. But this tried to squeeze the logic into CachingOptimizer, and tied CachingOptimizer and LazyBridgeOptimizer together too much.
A related PR is Concrete optimizer type in CachingOptimizer #2520. If JuMP's backend is not a concrete type, we don't need the CachingOptimizer to also be non-concrete.

Demo

By default, the LazyBridgeOptimizer is omitted. But, if you do something that requires bridges, they are added:

julia> model = Model(Clp.Optimizer)
A JuMP Model
Feasibility problem with:
Variables: 0
Model mode: AUTOMATIC
CachingOptimizer state: EMPTY_OPTIMIZER
Solver name: Clp

julia> backend(model)
MOIU.CachingOptimizer{Clp.Optimizer, MOIU.UniversalFallback{MOIU.Model{Float64}}}
in state EMPTY_OPTIMIZER
in mode AUTOMATIC
with model cache MOIU.UniversalFallback{MOIU.Model{Float64}}
  fallback for MOIU.Model{Float64}
with optimizer Clp.Optimizer

julia> @variable(model, x[1:4] in MOI.Nonnegatives(4))
4-element Vector{VariableRef}:
 x[1]
 x[2]
 x[3]
 x[4]

julia> backend(model)
MOIU.CachingOptimizer{MOIB.LazyBridgeOptimizer{MOIU.CachingOptimizer{Clp.Optimizer, MOIU.UniversalFallback{MOIU.Model{Float64}}}}, MOIU.UniversalFallback{MOIU.Model{Float64}}}
in state EMPTY_OPTIMIZER
in mode AUTOMATIC
with model cache MOIU.UniversalFallback{MOIU.Model{Float64}}
  fallback for MOIU.Model{Float64}
with optimizer MOIB.LazyBridgeOptimizer{MOIU.CachingOptimizer{Clp.Optimizer, MOIU.UniversalFallback{MOIU.Model{Float64}}}}
  with 0 variable bridges
  with 0 constraint bridges
  with 0 objective bridges
  with inner model MOIU.CachingOptimizer{Clp.Optimizer, MOIU.UniversalFallback{MOIU.Model{Float64}}}
    in state EMPTY_OPTIMIZER
    in mode AUTOMATIC
    with model cache MOIU.UniversalFallback{MOIU.Model{Float64}}
      fallback for MOIU.Model{Float64}
    with optimizer Clp.Optimizer

You can also pass force_bridge_formulation = true to force-add the bridges:

julia> model = Model(Clp.Optimizer; force_bridge_formulation = true)
A JuMP Model
Feasibility problem with:
Variables: 0
Model mode: AUTOMATIC
CachingOptimizer state: EMPTY_OPTIMIZER
Solver name: Clp

julia> backend(model)
MOIU.CachingOptimizer{MOIB.LazyBridgeOptimizer{MOIU.CachingOptimizer{Clp.Optimizer, MOIU.UniversalFallback{MOIU.Model{Float64}}}}, MOIU.UniversalFallback{MOIU.Model{Float64}}}
in state EMPTY_OPTIMIZER
in mode AUTOMATIC
with model cache MOIU.UniversalFallback{MOIU.Model{Float64}}
  fallback for MOIU.Model{Float64}
with optimizer MOIB.LazyBridgeOptimizer{MOIU.CachingOptimizer{Clp.Optimizer, MOIU.UniversalFallback{MOIU.Model{Float64}}}}
  with 0 variable bridges
  with 0 constraint bridges
  with 0 objective bridges
  with inner model MOIU.CachingOptimizer{Clp.Optimizer, MOIU.UniversalFallback{MOIU.Model{Float64}}}
    in state EMPTY_OPTIMIZER
    in mode AUTOMATIC
    with model cache MOIU.UniversalFallback{MOIU.Model{Float64}}
      fallback for MOIU.Model{Float64}
    with optimizer Clp.Optimizer

Benchmark

using JuMP

### No bridging

function foo(optimizer, bridge_formulation)
    model = Model(optimizer; force_bridge_formulation = force_bridge_formulation)
    set_silent(model)
    @variable(model, x >= 0)
    @constraint(model, 2x + 1 <= 1)
    @objective(model, Max, 1.0 * x)
    optimize!(model)
end

### ObjectiveFunction bridging

function foo_obj(optimizer)
    model = Model(optimizer)
    set_silent(model)
    @variable(model, x >= 0)
    @constraint(model, 2x + 1 <= 1)
    @objective(model, Max, x)
    optimize!(model)
end

### variable bridging

function foo_var(optimizer)
    model = Model(optimizer)
    set_silent(model)
    @variable(model, x[1:1] in MOI.Nonnegatives(1))
    @constraint(model, 2x[1] + 1 <= 1)
    @objective(model, Max, 1.0 * x[1])
    optimize!(model)
end

### constraint bridging

function foo_con(optimizer)
    model = Model(optimizer)
    set_silent(model)
    @variable(model, x[1:2])
    @constraint(model, x in MOI.Nonpositives(2))
    @objective(model, Max, 1.0 * x[1])
    optimize!(model)
end

import Clp
import GLPK
const optimizer = ARGS[1] == "clp" ? Clp.Optimizer : GLPK.Optimizer

if ARGS[2] == "--bridge"
    @eval @time foo(optimizer, true)
    @eval @time foo(optimizer, true)
elseif ARGS[2] == "--no-bridge"
    @eval @time foo(optimizer, false)
    @eval @time foo(optimizer, false)
elseif ARGS[2] == "--var"
    @eval @time foo_var(optimizer)
    @eval @time foo_var(optimizer)
elseif ARGS[2] == "--con"
    @eval @time foo_con(optimizer)
    @eval @time foo_con(optimizer)
else
    @assert ARGS[2] == "--obj"
    @eval @time foo_obj(optimizer)
    @eval @time foo_obj(optimizer)
end

Current master

(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl clp --bridge
 10.509123 seconds (25.20 M allocations: 1.570 GiB, 5.64% gc time, 21.39% compilation time)
  0.000679 seconds (1.36 k allocations: 83.797 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl clp --no-bridge
  4.220926 seconds (8.44 M allocations: 537.310 MiB, 3.65% gc time, 29.07% compilation time)
  0.000323 seconds (506 allocations: 42.453 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl clp --var      
 11.194838 seconds (26.03 M allocations: 1.645 GiB, 5.66% gc time, 21.81% compilation time)
  0.000734 seconds (1.59 k allocations: 99.672 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl clp --con
 11.305996 seconds (27.82 M allocations: 1.736 GiB, 5.79% gc time, 27.37% compilation time)
  0.000590 seconds (1.44 k allocations: 89.172 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl clp --obj
 11.233307 seconds (25.30 M allocations: 1.576 GiB, 5.60% gc time, 22.66% compilation time)
  0.000542 seconds (1.37 k allocations: 84.109 KiB)

(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl glpk --bridge
  6.395862 seconds (15.22 M allocations: 921.493 MiB, 4.76% gc time, 31.53% compilation time)
  0.000251 seconds (909 allocations: 45.906 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl glpk --no-bridge
  3.740884 seconds (5.39 M allocations: 318.667 MiB, 2.13% gc time, 99.96% compilation time)
  0.000174 seconds (358 allocations: 31.203 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl glpk --var      
  8.470519 seconds (17.92 M allocations: 1.082 GiB, 5.56% gc time, 27.10% compilation time)
  0.000546 seconds (1.14 k allocations: 61.703 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl glpk --con
  8.209768 seconds (19.46 M allocations: 1.158 GiB, 4.48% gc time, 36.46% compilation time)
  0.000514 seconds (1.06 k allocations: 55.734 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl glpk --obj
  6.853098 seconds (15.37 M allocations: 931.368 MiB, 4.19% gc time, 36.77% compilation time)
  0.000288 seconds (918 allocations: 46.203 KiB)

This PR

(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl clp --bridge
 10.589091 seconds (25.71 M allocations: 1.604 GiB, 6.78% gc time, 11.96% compilation time)
  0.000610 seconds (1.37 k allocations: 85.797 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl clp --no-bridge
  4.787131 seconds (9.11 M allocations: 577.229 MiB, 4.02% gc time, 27.56% compilation time)
  0.000388 seconds (522 allocations: 44.484 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl clp --var      
 12.136513 seconds (26.08 M allocations: 1.653 GiB, 5.59% gc time, 11.15% compilation time)
  0.000879 seconds (1.65 k allocations: 104.969 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl clp --con
 11.449779 seconds (27.80 M allocations: 1.740 GiB, 6.50% gc time, 15.22% compilation time)
  0.000612 seconds (1.47 k allocations: 93.453 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl clp --obj
 10.283270 seconds (25.37 M allocations: 1.583 GiB, 6.06% gc time, 23.26% compilation time)
  0.000537 seconds (1.40 k allocations: 88.359 KiB)

(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl glpk --bridge
  6.067656 seconds (15.35 M allocations: 932.123 MiB, 5.25% gc time, 22.02% compilation time)
  0.000261 seconds (926 allocations: 47.953 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl glpk --no-bridge
  3.772164 seconds (5.49 M allocations: 324.804 MiB, 2.41% gc time, 99.95% compilation time)
  0.000162 seconds (376 allocations: 33.266 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl glpk --var      
  7.915397 seconds (18.05 M allocations: 1.093 GiB, 5.46% gc time, 15.57% compilation time)
  0.000556 seconds (1.20 k allocations: 66.953 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl glpk --con
  7.842836 seconds (19.53 M allocations: 1.165 GiB, 4.78% gc time, 21.45% compilation time)
  0.000452 seconds (1.10 k allocations: 60.031 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. test/perf/force_bridge_formulation.jl glpk --obj
  7.193380 seconds (15.38 M allocations: 933.930 MiB, 4.45% gc time, 37.75% compilation time)
  0.000415 seconds (957 allocations: 50.484 KiB)

The big change is really that its 2x as fast when bridges are not needed.

Alternative

The alternative is to change the default of bridge_constraints to false, and also fix the double abstract-type issue.

This would give everyone a speed-up if they don't use bridges, and for models which use bridges, they can just go

Model(GLPK.Optimizer, bridge_formulation = true)

This would have the added benefit of giving people greater insight into when bridges get used.

TODO

TODOs before merging

Better tests

src/constraints.jl

src/optimizer_interface.jl

blegat · 2021-05-21T13:25:01Z

bridge_constraint does not have the same effect as before as setting it to false won't prevent bridges to be automatically added if needed.
If were going to be breaking, we might as well change the name since it should be changed before v1.0 (bridge_constraint does not make sense given that we also bridge variables and objectives).
It should be bridge_mode with

AUTO_BRIDGING: add if needed
ADD_BRIDGING: add from the start
NO_BRIDGING: don't add it even if needed

codecov · 2021-05-28T01:38:38Z

Codecov Report

Merging #2610 (e19f712) into master (b300820) will decrease coverage by 0.22%.
The diff coverage is 90.38%.

❗ Current head e19f712 differs from pull request most recent head d4d02b4. Consider uploading reports for the commit d4d02b4 to get more accurate results

@@            Coverage Diff             @@
##           master    #2610      +/-   ##
==========================================
- Coverage   94.18%   93.96%   -0.23%     
==========================================
  Files          43       43              
  Lines        5522     5564      +42     
==========================================
+ Hits         5201     5228      +27     
- Misses        321      336      +15

Impacted Files	Coverage Δ
src/variables.jl	`97.07% <85.36%> (-1.56%)`	⬇️
src/optimizer_interface.jl	`79.66% <86.66%> (+4.15%)`	⬆️
src/objective.jl	`94.00% <92.30%> (+0.38%)`	⬆️
src/JuMP.jl	`82.46% <96.29%> (-2.52%)`	⬇️
src/constraints.jl	`95.65% <100.00%> (+0.08%)`	⬆️
src/copy.jl	`95.31% <100.00%> (-0.08%)`	⬇️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b300820...d4d02b4. Read the comment docs.

odow · 2021-05-28T01:53:39Z

Docs are failing because of jump-dev/MathOptInterface.jl#1329. This shows that we're copying straight to GLPK, which should give a nice speedup though!

Alternate fix here: jump-dev/GLPK.jl#183

odow · 2021-09-27T04:05:23Z

Good news on the GLPK front, we're < 4 seconds, compared to 15 seconds in May.

(base) oscar@Oscars-MBP JuMP % ~/julia --project=. ~/Desktop/bench.jl glpk --bridge
  6.649749 seconds (15.36 M allocations: 933.000 MiB, 4.98% gc time, 21.35% compilation time)
  0.000296 seconds (926 allocations: 47.953 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. ~/Desktop/bench.jl glpk --no-bridge
  3.869593 seconds (5.49 M allocations: 324.722 MiB, 2.53% gc time, 99.96% compilation time)
  0.000163 seconds (376 allocations: 33.266 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. ~/Desktop/bench.jl glpk --var      
  7.808467 seconds (18.06 M allocations: 1.094 GiB, 5.11% gc time, 15.47% compilation time)
  0.000547 seconds (1.20 k allocations: 66.953 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. ~/Desktop/bench.jl glpk --con
  8.049347 seconds (19.54 M allocations: 1.166 GiB, 4.39% gc time, 20.44% compilation time)
  0.000466 seconds (1.10 k allocations: 60.031 KiB)
(base) oscar@Oscars-MBP JuMP % ~/julia --project=. ~/Desktop/bench.jl glpk --obj
  6.487435 seconds (15.40 M allocations: 934.871 MiB, 4.42% gc time, 36.21% compilation time)
  0.000282 seconds (957 allocations: 50.484 KiB)

odow · 2021-10-03T23:19:58Z

~~I need to rerun the benchmark because I don't think I precompiled things properly, but this is the sort of thing our new benchmarking scripts can produce:~~

time_A is the current master
time_B is this PR
time_ratio is time_B / time_A (smaller is better)

43×4 DataFrame
 Row │ models                             time_A    time_B    time_ratio 
     │ String                             Float64   Float64   Float64    
─────┼───────────────────────────────────────────────────────────────────
   1 │ getting_started_with_JuMP.jl       15.6854   12.0367     0.767382
   2 │ solvers_and_solutions.jl           14.7179   11.8145     0.80273
   3 │ diet.jl                            14.7689   12.3716     0.837675
   4 │ cannery.jl                         14.6897   12.3197     0.838661
   5 │ urban_plan.jl                      16.2603   13.6643     0.840345
   6 │ knapsack.jl                        13.3857   11.5781     0.864959
   7 │ factory_schedule.jl                17.572    15.2579     0.86831
   8 │ transp.jl                          14.4346   12.6629     0.877258
   9 │ multi.jl                           14.6595   12.9403     0.882725
  10 │ cutting_stock_column_generation.…  17.9306   15.8623     0.88465
  11 │ network_flows.jl                   15.9198   14.1367     0.887992
  12 │ variables_constraints_objective.…  20.2857   18.0255     0.888579
  13 │ sudoku.jl                          13.568    12.078      0.890187
  14 │ n-queens.jl                        13.3915   11.949      0.892282
  15 │ working_with_data_files.jl         40.6386   36.3744     0.89507
  16 │ power_systems.jl                   19.587    17.5443     0.895711
  17 │ performance_tips.jl                 7.95099   7.17228    0.902061
  18 │ steelT3.jl                         18.3476   16.5523     0.902149
  19 │ facility_location.jl               25.7838   23.3619     0.906069
  20 │ clnlbeam.jl                        24.7641   23.0942     0.932567
  21 │ nlp_tricks.jl                      23.291    21.8562     0.938394
  22 │ prod.jl                            18.1101   17.0073     0.939106
  23 │ callbacks.jl                       18.1412   17.3113     0.954257
  24 │ qcp.jl                             23.407    22.6249     0.966589
  25 │ rosenbrock.jl                      21.1796   20.7913     0.981668
  26 │ rocket_control.jl                  36.0933   35.5357     0.984552
  27 │ geographic_clustering.jl           25.5792   25.288      0.988618
  28 │ mle.jl                             23.1395   23.0961     0.998125
  29 │ an_introduction_to_julia.jl         7.65756   7.6937     1.00472
  30 │ portfolio.jl                       23.3081   23.4615     1.00658
  31 │ corr_sdp.jl                        24.8389   25.7969     1.03857
  32 │ min_ellipse.jl                     23.7421   24.7362     1.04187
  33 │ logistic_regression.jl             30.0491   31.5174     1.04886
  34 │ min_distortion.jl                  24.5027   25.7974     1.05284
  35 │ space_shuttle_reentry_trajectory…  41.9092   44.1962     1.05457
  36 │ max_cut_sdp.jl                     25.1358   26.6172     1.05894
  37 │ experiment_design.jl               27.9409   29.5918     1.05908
  38 │ tips_and_tricks.jl                 12.0369   13.018      1.08151
  39 │ finance.jl                         15.8194   17.2735     1.09192
  40 │ robust_uncertainty.jl              26.4479   29.0361     1.09786
  41 │ cluster.jl                         25.9001   28.4743     1.09939
  42 │ benders_lazy_constraints.jl        17.6766   21.025      1.18942
  43 │ benders_decomposition.jl           17.329    20.8819     1.20502

src/JuMP.jl

src/optimizer_interface.jl

blegat · 2021-10-08T18:56:39Z

src/optimizer_interface.jl

@@ -186,6 +205,16 @@ function optimize!(
                "The solver does not support nonlinear problems " *
                "(i.e., NLobjective and NLconstraint).",
            )
+        elseif err isa MOI.UnsupportedConstraint
+            if _add_bridges_if_needed(model)


Shouldn't we also catch unsupported objective function ?

Maybe also free variables that are not supported ? You should probably try with CSDP see the error that is thrown

blegat · 2021-10-08T19:04:27Z

src/JuMP.jl

+    end
+
+    model = direct_model(
+        MOI.Utilities.CachingOptimizer(


Here, we could directly do direct_model(MOIU.UniversalFallback(MOIU.Model{Float64}())). Otherwise we will compile a lot of methods for MOIU.CachingOptimizer{...,MOI.AbstractOptimizer} but we will never use the fact that it's a caching optimizer. If we use direct_model(MOIU.UniversalFallback(MOIU.Model{Float64}())) directly, wouldn't we gain in compile time ?
It won't add more possible states. Currently the possible states are:

The MOI backend is a CachingOptimizer in state NO_OPTIMIZER

An optimizer is set
and if we do the change I suggest, the possible states are

The MOI backend is UniversalFallback{Model}

An optimizer is set and the MOI backend is a CachingOptimizer

This doesn't work for two reasons:

UniversalFallback is not an AbstractOptimizer

We wouldn't be able to tell if the user was in direct mode

Rebase and fix formatting bridge_formulation instead of bridge_constraints Some test fixes More fixes Fix docs. Requires new GLPK release Update models.md Remove TODO Another fix More fixes Fix formatting Fix typo More tests Update solvers_and_solutions.jl Rename bridge_formulation to force_bridge_formulation

odow · 2021-10-14T22:39:24Z

I'm having second thoughts about this PR. Perhaps we just need to:

Fix the double-abstract-type issue (Concrete optimizer type in CachingOptimizer #2520)
Make the documentation much clearer and explicit that people should try bridge_constraints = false to get a speed boost
Rename bridge_constraints to add_bridges

Model(GLPK.Optimizer; add_bridges = false) isn't that bad. It's explicit in what it is doing, and it doesn't rely on this hacky trying to catch errors mechanism.

odow added the Type: Performance label May 21, 2021

odow mentioned this pull request May 21, 2021

WIP: only bridge if needed #2513

Closed

odow added the breaking label May 21, 2021

blegat reviewed May 21, 2021

View reviewed changes

src/constraints.jl Show resolved Hide resolved

blegat reviewed May 21, 2021

View reviewed changes

src/optimizer_interface.jl Outdated Show resolved Hide resolved

This was referenced May 27, 2021

Use unsafe_backend in various places #2615

Merged

Rename backend variables to moi_backend #2616

Merged

odow force-pushed the od/auto-bridge branch from e5e0918 to 84fb708 Compare May 27, 2021 23:40

This was referenced May 28, 2021

Fix index_map in MOI.copy_to jump-dev/GLPK.jl#183

Merged

Rewrite MOI.copy_to to simplify and fix various bugs jump-dev/GLPK.jl#184

Merged

blegat mentioned this pull request Jun 19, 2021

Simplify handling of linear and quadratic derivatives for nonlinear solvers jump-dev/MathOptInterface.jl#1397

Closed

odow mentioned this pull request Jun 21, 2021

Skip the inner (inner) cache if bridges are not used jump-dev/MathOptInterface.jl#1251

Closed

odow force-pushed the od/auto-bridge branch from cc43c04 to d2f253c Compare September 27, 2021 02:20

This was referenced Sep 27, 2021

Parametrize JuMP model in optimizer type #1348

Closed

Make some moi_xxx functions private #2548

Closed

odow force-pushed the od/auto-bridge branch from c5283eb to b4c8f83 Compare September 29, 2021 03:56

odow changed the title ~~WIP: bridge backend of Model if needed~~ Bridge backend of Model if needed Sep 29, 2021

odow force-pushed the od/auto-bridge branch 2 times, most recently from 2569bf7 to 9c04228 Compare October 1, 2021 00:50

odow force-pushed the od/auto-bridge branch 2 times, most recently from 22f844d to ad23255 Compare October 4, 2021 21:33

odow added the Status: Needs developer call This should be discussed on a monthly developer call label Oct 7, 2021

odow requested a review from blegat October 7, 2021 21:30

blegat reviewed Oct 8, 2021

View reviewed changes

src/JuMP.jl Outdated Show resolved Hide resolved

blegat reviewed Oct 8, 2021

View reviewed changes

src/JuMP.jl Outdated Show resolved Hide resolved

blegat reviewed Oct 8, 2021

View reviewed changes

src/optimizer_interface.jl Outdated Show resolved Hide resolved

blegat reviewed Oct 8, 2021

View reviewed changes

odow added 4 commits October 15, 2021 11:28

Add benchmark script

5f4205a

Update benchmarks

1568952

Fix formatting

e19f712

odow force-pushed the od/auto-bridge branch from 4bf6424 to e19f712 Compare October 14, 2021 22:28

Fix comments

d4d02b4

odow mentioned this pull request Oct 15, 2021

[breaking] Make moi_backend a concrete type and remove MANUAL mode #2756

Merged

odow closed this in #2756 Oct 20, 2021

odow deleted the od/auto-bridge branch November 10, 2021 02:28

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Bridge backend of Model if needed #2610

Bridge backend of Model if needed #2610

odow commented May 21, 2021 •

edited

Loading

blegat commented May 21, 2021

codecov bot commented May 28, 2021 •

edited

Loading

odow commented May 28, 2021 •

edited

Loading

odow commented Sep 27, 2021

odow commented Oct 3, 2021 •

edited

Loading

blegat Oct 8, 2021

blegat Oct 8, 2021

blegat Oct 8, 2021

odow Oct 14, 2021

odow commented Oct 14, 2021

Bridge backend of Model if needed #2610

Bridge backend of Model if needed #2610

Conversation

odow commented May 21, 2021 • edited Loading

Closes

Background

Demo

Benchmark

Alternative

TODO

blegat commented May 21, 2021

codecov bot commented May 28, 2021 • edited Loading

Codecov Report

odow commented May 28, 2021 • edited Loading

odow commented Sep 27, 2021

odow commented Oct 3, 2021 • edited Loading

blegat Oct 8, 2021

Choose a reason for hiding this comment

blegat Oct 8, 2021

Choose a reason for hiding this comment

blegat Oct 8, 2021

Choose a reason for hiding this comment

odow Oct 14, 2021

Choose a reason for hiding this comment

odow commented Oct 14, 2021

odow commented May 21, 2021 •

edited

Loading

codecov bot commented May 28, 2021 •

edited

Loading

odow commented May 28, 2021 •

edited

Loading

odow commented Oct 3, 2021 •

edited

Loading