Skip to content

Commit

Permalink
Merge b86460e into 67133a8
Browse files Browse the repository at this point in the history
  • Loading branch information
@IainNZ
IainNZ committed Nov 10, 2018
2 parents 67133a8 + b86460e commit d7cf8e4
Showing 1 changed file with 87 additions and 139 deletions.
226 changes: 87 additions & 139 deletions test/perf/speed.jl
View file
@@ -1,102 +1,83 @@
# Copyright 2017, Iain Dunning, Joey Huchette, Miles Lubin, and contributors
# This Source Code Form is subject to the terms of the Mozilla Public
# License, v. 2.0. If a copy of the MPL was not distributed with this
# file, You can obtain one at http://mozilla.org/MPL/2.0/.
#############################################################################
# JuMP
# An algebraic modelling language for Julia
# See http://github.com/JuliaOpt/JuMP.jl
#############################################################################
# speed.jl
#
# Runs some JuMP benchmarks to test for speed-related regressions.
# Examples taken from the Lubin, Dunning paper.
# Benchmarks model building time to test for performance regressions.
# Based on the models benchmarked in the paper:
# Lubin, M., & Dunning, I. (2015).
# Computing in operations research using Julia.
# INFORMS Journal on Computing, 27(2), 238-248.
#
# Post past results here
# ---------------------------------------------------------------------------
# Iain's Dell Laptop
# 2013/10/05 ee443c8d6a779ff74373178dbd6086b1912c5f5e
# Run 1
# PMEDIAN BUILD MIN=0.37519394 MED=0.72586158
# PMEDIAN WRITE MIN=1.551488986 MED=1.62851847
# CONT5 BUILD MIN=0.254537676 MED=0.454625898
# CONT5 WRITE MIN=1.581302132 MED=1.639451427
# Run 2
# PMEDIAN BUILD MIN=0.381689249 MED=0.745308535
# PMEDIAN WRITE MIN=1.55169788 MED=1.580791808
# CONT5 BUILD MIN=0.248255523 MED=0.454527841
# CONT5 WRITE MIN=1.60369395 MED=1.638122135
# 2013/11/10 042144bf304af0c6f672c87040edf3cffa5890a3
# Run 1
# PMEDIAN BUILD MIN=0.386472824 MED=0.577460239
# PMEDIAN WRITE MIN=1.487720674 MED=1.573461681
# CONT5 BUILD MIN=0.237566804 MED=0.43350331
# CONT5 WRITE MIN=1.554705597 MED=1.566282347
# ---------------------------------------------------------------------------
# Iain's Desktop
# 2013/10/23 d8c64fd341801a5c266597df4ec52377f42a5260
# Run 1
# PMEDIAN BUILD MIN=0.267263966 MED=0.272618701
# PMEDIAN WRITE MIN=1.263382472 MED=1.26997836
# CONT5 BUILD MIN=0.118468756 MED=0.18796253
# CONT5 WRITE MIN=1.309088036 MED=1.325527758
# Run 2
# PMEDIAN BUILD MIN=0.271933082 MED=0.274442777
# PMEDIAN WRITE MIN=1.29212427 MED=1.30086527
# CONT5 BUILD MIN=0.123911798 MED=0.193959468
# CONT5 WRITE MIN=1.293636495 MED=1.305372909
# 2014/02/09 8e96ed879a58134207e4d61dc03766bd67d72523
# PMEDIAN BUILD MIN=0.199827121 MED=0.309448725
# PMEDIAN WRITE MIN=1.159296703 MED=1.167277902
# CONT5 BUILD MIN=0.12434096 MED=0.19083312
# CONT5 WRITE MIN=1.142068783 MED=1.149582037
#############################################################################

using JuMP
import BenchmarkTools: @benchmark, allocs
using Compat
using Compat.Random
@static if VERSION >= v"0.7.0-DEV.3406"
srand(seed) = Random.seed!(seed)
end

function pMedian(numFacility::Int,numCustomer::Int,numLocation::Int,useMPS)
srand(10)
customerLocations = [rand(1:numLocation) for a = 1:numCustomer ]

buildTime = @elapsed begin
m = Model()
using JuMP

# Facility locations
@variable(m, 0 <= s[1:numLocation] <= 1)
"""
p_median(num_facility, num_customer, num_location)
Implements the "p-median" facility location problem. We try to locate N
facilities such that we minimize the distance any given customer has to travel
to reach their closest facility. In this simple instance we will operate
in a 1D world with L possible locations for facilities, and customers being
located at random locations along the number line from 1 to D.
We use anonymous variables to remove the cost of name generation from the
benchmark.
"""
function p_median(num_facilities, num_customers, num_locations)
Random.seed!(10)
customer_locations = [rand(1:num_locations) for _ in 1:num_customers]

model = Model()
has_facility = @variable(model, [1:num_locations], Bin)
is_closest = @variable(model, [1:num_locations, 1:num_customers], Bin)

@objective(model, Min,
sum(abs(customer_locations[customer] - location)
* is_closest[location, customer]
for customer in 1:num_customers, location in 1:num_locations))

for customer in 1:num_customers
# `location` can't be closest for `customer` if there is no facility.
@constraint(model,
[location in 1:num_locations],
is_closest[location, customer] <= has_facility[location])
# One facility must be the closest for `customer`.
@constraint(model,
sum(is_closest[location, customer]
for location in 1:num_locations) == 1)
end

# Aux. variable: x_a,i = 1 iff the closest facility to a is at i
@variable(m, 0 <= x[1:numLocation,1:numCustomer] <= 1)
# Must place all facilities.
@constraint(model, sum(has_facility) == num_facilities)
end

# Objective: min distance
@objective(m, Max, sum(abs(customerLocations[a]-i)*x[i,a] for a = 1:numCustomer, i = 1:numLocation) )
println("P-Median(100 facilities, 100 customers, 5000 locations) benchmark:")
result = @benchmark p_median(100, 100, 5000)
display(result)
println()

# Constraints
for a in 1:numCustomer
# Subject to linking x with s
for i in 1:numLocation
@constraint(m, x[i,a] - s[i] <= 0)
end
# Subject to one of x must be 1
@constraint(m, sum(x[i,a] for i=1:numLocation) == 1 )
end

# Subject to must allocate all facilities
@constraint(m, sum(s[i] for i=1:numLocation) == numFacility )
end
"""
cont5(n)
writeTime = @elapsed begin
if useMPS
writeMPS(m,"/dev/null")
else
writeLP(m,"/dev/null")
end
end
Based on a linear-Quadratic control problem (cont5_2_1) from one of Hans
Mittleman's instance collections.
return buildTime, writeTime
end

function cont5(n,useMPS)
We use anonymous variables to remove the cost of name generation from the
benchmark.
"""
function cont5(n)
m = n
n1 = n-1
m1 = m-1
Expand All @@ -105,72 +86,39 @@ function cont5(n,useMPS)
dt = T/m
h2 = dx^2
a = 0.001
yt = [0.5*(1 - (j*dx)^2) for j=0:n]

buildTime = @elapsed begin
mod = Model()
@variable(mod, 0 <= y[0:m,0:n] <= 1)
@variable(mod, -1 <= u[1:m] <= 1)
@objective(mod, Min, y[0,0])

# PDE
for i = 0:m1
for j = 1:n1
@constraint(mod, h2*(y[i+1,j] - y[i,j]) == 0.5*dt*(y[i,j-1] - 2*y[i,j] + y[i,j+1] + y[i+1,j-1] - 2*y[i+1,j] + y[i+1,j+1]) )
end
end
yt = [0.5*(1 - (j*dx)^2) for j in 0:n]

# IC
for j = 0:n
@constraint(mod, y[0,j] == 0)
end
model = Model()
y = @variable(model, [0:m,0:n], lower_bound=0, upper_bound=1)
u = @variable(model, [1:m], lower_bound=-1, upper_bound=1)

# BC
for i = 1:m
@constraint(mod, y[i,2] - 4*y[i,1] + 3*y[i,0] == 0)
@constraint(mod, y[i,n-2] - 4*y[i,n1] + 3*y[i,n] == (2*dx)*(u[i] - y[i,n]))
end
end
@objective(model, Min, y[0,0])

writeTime = @elapsed begin
if !useMPS
writeLP(mod, "/dev/null")
else
writeMPS(mod, "/dev/null")
# PDE
for i in 0:m1
for j in 1:n1
@constraint(model,
h2*(y[i+1,j] - y[i,j])
== 0.5*dt*(y[i,j-1] - 2*y[i,j] +
y[i,j+1] + y[i+1,j-1] - 2*y[i+1,j] + y[i+1,j+1]) )
end
end

return buildTime, writeTime
end


function RunTests()
# Pmedian
pmedian_build = Float64[]
pmedian_write = Float64[]
for runs = 1:9
bt, wt = pMedian(100,100,5000,false)
push!(pmedian_build, bt)
push!(pmedian_write, wt)
# Initial conditions.
for j in 0:n
@constraint(model, y[0,j] == 0)
end
sort!(pmedian_build)
sort!(pmedian_write)
print("PMEDIAN BUILD MIN=",minimum(pmedian_build)," MED=",pmedian_build[5],"\n")
print("PMEDIAN WRITE MIN=",minimum(pmedian_write)," MED=",pmedian_write[5],"\n")

# Cont5
cont5_build = Float64[]
cont5_write = Float64[]
for runs = 1:9
bt, wt = cont5(500,false)
push!(cont5_build, bt)
push!(cont5_write, wt)
end
sort!(cont5_build)
sort!(cont5_write)
print("CONT5 BUILD MIN=",minimum(cont5_build)," MED=",cont5_build[5],"\n")
print("CONT5 WRITE MIN=",minimum(cont5_write)," MED=",cont5_write[5],"\n")

# Boundary conditions.
for i in 1:m
@constraint(model,
y[i,2] - 4*y[i,1] + 3*y[i,0] == 0)
@constraint(model,
y[i,n-2] - 4*y[i,n1] + 3*y[i,n]== (2*dx)*(u[i] - y[i,n]))
end
end

RunTests()
println("Cont5(n=500) benchmark:")
result = @benchmark cont5(500)
display(result)
println()

0 comments on commit d7cf8e4

Please sign in to comment.