kalmar
/
PropertyT.jl
mirror of https://github.com/kalmarek/PropertyT.jl.git
1
0
Fork 0
Code Issues Releases Wiki Activity

Merge branch 'enh/rework-logging' of https://git.wmi.amu.edu.pl/kalmar/PropertyT.jl into enh/rework-logging

This commit is contained in:
kalmarek 2018-01-04 21:46:00 +01:00
parent 596b688c1f cb3c6a0ef5
commit 8413254917
6 changed files with 680 additions and 706 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

82
src/CheckSolution.jl
View File

@ -36,9 +36,8 @@ function compute_SOS(Q::AbstractArray, pm::Array{Int,2}, l::Int)
groupring_square(Q[:,i], l, pm)
end
println("")
return result
end
function compute_SOS(Q::AbstractArray, RG::GroupRing, l::Int)
@ -46,32 +45,13 @@ function compute_SOS(Q::AbstractArray, RG::GroupRing, l::Int)
return GroupRingElem(result, RG)
end
function distance_to_cone{T<:Interval}(elt::GroupRingElem, Q::AbstractArray{T,2}, wlen::Int)
SOS = compute_SOS(Q, parent(elt), length(elt.coeffs))
SOS_diff = elt - SOS
function distances_to_cone(elt::GroupRingElem, wlen::Int)
ɛ_dist = GroupRings.augmentation(elt)
ɛ_dist = GroupRings.augmentation(SOS_diff)
info(logger, "ɛ(∑ξᵢ*ξᵢ) ∈ $(ɛ_dist)")
eoi_SOS_L1_dist = norm(SOS_diff,1)
info(logger, "‖Δ² - λΔ - ∑ξᵢ*ξᵢ‖₁ ∈ $(eoi_SOS_L1_dist)")
eoi_SOS_L1_dist = norm(elt,1)
dist = 2^(wlen-1)*eoi_SOS_L1_dist
return dist
end
function distance_to_cone{T}(elt::GroupRingElem, Q::AbstractArray{T,2}, wlen::Int)
SOS = compute_SOS(Q, parent(elt), length(elt.coeffs))
SOS_diff = elt - SOS
ɛ_dist = GroupRings.augmentation(SOS_diff)
info(logger, "ɛ(Δ² - λΔ - ∑ξᵢ*ξᵢ) ≈ $(@sprintf("%.10f", ɛ_dist))")
eoi_SOS_L1_dist = norm(SOS_diff,1)
info(logger, "‖Δ² - λΔ - ∑ξᵢ*ξᵢ‖₁ ≈ $(@sprintf("%.10f", eoi_SOS_L1_dist))")
dist = 2^(wlen-1)*eoi_SOS_L1_dist
return dist
return dist, ɛ_dist, eoi_SOS_L1_dist
end
function augIdproj{T, I<:AbstractInterval}(S::Type{I}, Q::AbstractArray{T,2})
@ -99,30 +79,56 @@ function augIdproj{T}(Q::AbstractArray{T,2}, logger)
return Q
end
function distance_to_positive_cone(Δ::GroupRingElem, λ, Q, wlen::Int)
function distance_to_cone(elt::GroupRingElem, λ::T, Q::AbstractArray{T,2}, wlen::Int, logger) where {T<:AbstractFloat}
info(logger, "------------------------------------------------------------")
info(logger, "λ = $λ")
info(logger, "Checking in floating-point arithmetic...")
Δ²_λΔ = EOI(Δ, λ)
@logtime logger fp_distance = λ - distance_to_cone(Δ²_λΔ, Q, wlen)
info(logger, "Floating point distance (to positive cone) ≈ $(@sprintf("%.10f", fp_distance))")
@logtime logger SOS_diff = elt - compute_SOS(Q, parent(elt), length(elt.coeffs))
dist, ɛ_dist, eoi_SOS_L1_dist = distances_to_cone(SOS_diff, wlen)
info(logger, "ɛ(Δ² - λΔ - ∑ξᵢ*ξᵢ) ≈ $(@sprintf("%.10f", ɛ_dist))")
info(logger, "‖Δ² - λΔ - ∑ξᵢ*ξᵢ‖₁ ≈ $(@sprintf("%.10f", eoi_SOS_L1_dist))")
fp_distance = λ - dist
info(logger, "Floating point distance (to positive cone) ≈")
info(logger, "$(@sprintf("%.10f", fp_distance))")
info(logger, "")
return fp_distance
end
function distance_to_cone(elt::GroupRingElem, λ::T, Q::AbstractArray{T,2}, wlen::Int, logger) where {T<:AbstractInterval}
info(logger, "------------------------------------------------------------")
info(logger, "λ = $λ")
info(logger, "Checking in interval arithmetic...")
@logtime logger SOS_diff = elt - compute_SOS(Q, parent(elt), length(elt.coeffs))
dist, ɛ_dist, eoi_SOS_L1_dist = distances_to_cone(SOS_diff, wlen)
info(logger, "ɛ(∑ξᵢ*ξᵢ) ∈ $(ɛ_dist)")
info(logger, "‖Δ² - λΔ - ∑ξᵢ*ξᵢ‖₁ ∈ $(eoi_SOS_L1_dist)")
int_distance = λ - dist
info(logger, "The Augmentation-projected actual distance (to positive cone) ∈")
info(logger, "$(int_distance)")
info(logger, "")
return int_distance
end
function check_distance_to_cone(Δ::GroupRingElem, λ, Q, wlen::Int, logger)
fp_distance = distance_to_cone(EOI(Δ, λ), λ, Q, wlen, logger)
if fp_distance ≤ 0
return fp_distance
end
info(logger, "")
Q = augIdproj(Q, logger)
info(logger, "Checking in interval arithmetic")
λ = @interval(λ)
Δ = GroupRingElem([@interval(c) for c in Δ.coeffs], parent(Δ))
Δ²_λΔ = EOI(Δ, λ)
Q = augIdproj(Q, logger)
@logtime logger Interval_dist_to_ΣSq = λ - distance_to_cone(Δ²_λΔ, Q, wlen)
info(logger, "The Augmentation-projected actual distance (to positive cone) ∈ $(Interval_dist_to_ΣSq)")
info(logger, "------------------------------------------------------------")
int_distance = distance_to_cone(EOI(Δ, λ), λ, Q, wlen, logger)
return Interval_dist_to_ΣSq
return int_distance.lo
end

100
src/Orbit-wise.jl
View File

@ -14,6 +14,7 @@ immutable Settings{T<:AbstractMathProgSolver}
upper_bound::Float64
tol::Float64
warmstart::Bool
logger
end
prefix(s::Settings) = s.name
@ -32,23 +33,23 @@ immutable OrbitData{T<:AbstractArray{Float64, 2}, LapType <:AbstractVector{Float
end
function OrbitData(sett::Settings)
splap = load(joinpath(prepath(sett), "delta.jld"), "Δ");
pm = load(joinpath(prepath(sett), "pm.jld"), "pm");
splap = load(filename(prepath(sett), :Δ), "Δ");
pm = load(filename(prepath(sett), :pm), "pm");
cnstr = PropertyT.constraints(pm);
splap² = similar(splap)
splap² = GroupRings.mul!(splap², splap, splap, pm);
Uπs = load(joinpath(prepath(sett), "U_pis.jld"), "Uπs")
Uπs = load(filename(prepath(sett), :Uπs), "Uπs")
nzros = [i for i in 1:length(Uπs) if size(Uπs[i],2) !=0]
Uπs = Uπs[nzros]
Uπs = sparsify!.(Uπs, sett.tol, check=true, verbose=true)
#dimensions of the corresponding πs:
dims = load(joinpath(prepath(sett), "U_pis.jld"), "dims")[nzros]
dims = load(filename(prepath(sett), :Uπs), "dims")[nzros]
m, P = init_model(size(Uπs,1), [size(U,2) for U in Uπs]);
orbits = load(joinpath(prepath(sett), "orbits.jld"), "orbits");
orbits = load(filename(prepath(sett), :orb), "orbits");
n = size(Uπs[1],1)
orb_spcnstrm = [orbit_constraint(cnstr[collect(orb)], n) for orb in orbits]
orb_splap = orbit_spvector(splap, orbits)
@ -79,7 +80,7 @@ function sparsify!{Tv,Ti}(M::SparseMatrixCSC{Tv,Ti}, eps=eps(Tv); verbose=false)
m = nnz(M)
if verbose
info(logger, "Sparsified density:", rpad(densM, 20), " → ", rpad(dens(M), 20))
info("Sparsified density:", rpad(densM, 20), " → ", rpad(dens(M), 20))
end
return M
@ -91,11 +92,11 @@ function sparsify!{T}(M::AbstractArray{T}, eps=eps(T); check=false, verbose=fals
M[abs.(M) .< eps] .= zero(T)
if check && rankM != rank(M)
warn(logger, "Sparsification decreased the rank!")
warn("Sparsification decreased the rank!")
end
if verbose
info(logger, "Sparsified density:", rpad(densM, 20), " → ", rpad(dens(M),20))
info("Sparsified density:", rpad(densM, 20), " → ", rpad(dens(M),20))
end
return sparse(M)
@ -103,19 +104,6 @@ end
sparsify{T}(U::AbstractArray{T}, tol=eps(T); check=true, verbose=false) = sparsify!(deepcopy(U), tol, check=check, verbose=verbose)
function init_orbit_data(logger, sett::Settings; radius=2)
ex(fname) = isfile(joinpath(prepath(sett), fname))
files_exists = ex.(["delta.jld", "pm.jld", "U_pis.jld", "orbits.jld", "preps.jld"])
if !all(files_exists)
compute_orbit_data(logger, prepath(sett), sett.G, sett.S, sett.autS, radius=radius)
end
return 0
end
function transform(U::AbstractArray, V::AbstractArray; sparse=true)
if sparse
return sparsify!(U'*V*U)
@ -178,8 +166,8 @@ function init_model(n, sizes)
end
function create_SDP_problem(sett::Settings)
info(logger, "Loading orbit data....")
@logtime logger SDP_problem, orb_data = OrbitData(sett);
info(sett.logger, "Loading orbit data....")
@logtime sett.logger SDP_problem, orb_data = OrbitData(sett);
if sett.upper_bound < Inf
λ = JuMP.getvariable(SDP_problem, :λ)
@ -187,8 +175,8 @@ function create_SDP_problem(sett::Settings)
end
t = length(orb_data.laplacian)
info(logger, "Adding $t constraints ... ")
@logtime logger addconstraints!(SDP_problem, orb_data)
info(sett.logger, "Adding $t constraints ... ")
@logtime sett.logger addconstraints!(SDP_problem, orb_data)
return SDP_problem, orb_data
end
@ -201,16 +189,16 @@ function λandP(m::JuMP.Model, data::OrbitData, warmstart=true)
end
function λandP(m::JuMP.Model, data::OrbitData, sett::Settings)
info(logger, "Solving SDP problem...")
λ, Ps = λandP(m, data, sett.warmstart)
info(sett.logger, "Solving SDP problem...")
@logtime sett.logger λ, Ps = λandP(m, data, sett.warmstart)
info(logger, "Reconstructing P...")
info(sett.logger, "Reconstructing P...")
preps = load_preps(joinpath(prepath(sett), "preps.jld"), sett.autS)
preps = load_preps(filename(prepath(sett), :preps), sett.autS)
@logtime logger recP = reconstruct_sol(preps, data.Us, Ps, data.dims)
@logtime sett.logger recP = reconstruct_sol(preps, data.Us, Ps, data.dims)
fname = PropertyT.λSDPfilenames(fullpath(sett))[2]
fname = filename(fullpath(sett), :P)
save(fname, "origP", Ps, "P", recP)
return λ, recP
end
@ -229,49 +217,39 @@ end
function check_property_T(sett::Settings)
init_orbit_data(logger, sett, radius=sett.radius)
ex(s) = exists(filename(prepath(sett), s))
if !sett.warmstart && all(isfile.(λSDPfilenames(fullpath(sett))))
files_exists = ex.([:pm, :Δ, :Uπs, :orb, :preps])
if !all(files_exists)
compute_orbit_data(sett.logger, prepath(sett), sett.S, sett.autS, radius=sett.radius)
end
cond1 = exists(filename(fullpath(sett), :λ))
cond2 = exists(filename(fullpath(sett), :P))
if !sett.warmstart && cond1 && cond2
λ, P = PropertyT.λandP(fullpath(sett))
else
info(logger, "Creating SDP problem...")
info(sett.logger, "Creating SDP problem...")
SDP_problem, orb_data = create_SDP_problem(sett)
JuMP.setsolver(SDP_problem, sett.solver)
info(sett.logger, Base.repr(SDP_problem))
λ, P = λandP(SDP_problem, orb_data, sett)
end
info(logger, "λ = $λ")
info(logger, "sum(P) = $(sum(P))")
info(logger, "maximum(P) = $(maximum(P))")
info(logger, "minimum(P) = $(minimum(P))")
if λ > 0
pm_fname, Δ_fname = pmΔfilenames(prepath(sett))
RG = GroupRing(sett.G, load(pm_fname, "pm"))
Δ = GroupRingElem(load(Δ_fname, "Δ")[:, 1], RG)
info(sett.logger, "λ = $λ")
info(sett.logger, "sum(P) = $(sum(P))")
info(sett.logger, "maximum(P) = $(maximum(P))")
info(sett.logger, "minimum(P) = $(minimum(P))")
isapprox(eigvals(P), abs.(eigvals(P)), atol=sett.tol) ||
warn("The solution matrix doesn't seem to be positive definite!")
# @assert P == Symmetric(P)
@logtime logger Q = real(sqrtm(Symmetric(P)))
sgap = distance_to_positive_cone(Δ, λ, Q, 2*sett.radius)
if isa(sgap, Interval)
sgap = sgap.lo
if λ > 0
return check_λ(sett.name, sett.S, λ, P, sett.radius, sett.logger)
end
if sgap > 0
info(logger, "λ ≥ $(Float64(trunc(sgap,12)))")
Kazhdan_κ = PropertyT.Kazhdan_from_sgap(sgap, length(sett.S))
Kazhdan_κ = Float64(trunc(Kazhdan_κ, 12))
info(logger, "κ($(sett.name), S) ≥ $Kazhdan_κ: Group HAS property (T)!")
return true
else
sgap = Float64(trunc(sgap, 12))
info(logger, "λ($(sett.name), S) ≥ $sgap: Group may NOT HAVE property (T)!")
return false
end
end
info(logger, "κ($(sett.name), S) ≥ $λ < 0: Tells us nothing about property (T)")
info(sett.logger, "κ($(sett.name), S) ≥ $λ < 0: Tells us nothing about property (T)")
return false
end

11
src/OrbitDecomposition.jl
View File

@ -166,12 +166,13 @@ function orthSVD{T}(M::AbstractMatrix{T})
return fact[:U][:,1:M_rank]
end
function compute_orbit_data{T<:GroupElem}(logger, name::String, G::Nemo.Group, S::Vector{T}, autS::Nemo.Group; radius=2)
function compute_orbit_data{T<:GroupElem}(logger, name::String, S::Vector{T}, autS::Nemo.Group; radius=2)
isdir(name) || mkdir(name)
info(logger, "Generating ball of radius $(2*radius)")
# TODO: Fix that by multiple dispatch?
G = parent(first(S))
Id = (isa(G, Nemo.Ring) ? one(G) : G())
@logtime logger E_2R, sizes = Groups.generate_balls(S, Id, radius=2*radius);
@ -182,10 +183,10 @@ function compute_orbit_data{T<:GroupElem}(logger, name::String, G::Nemo.Group, S
info(logger, "Product matrix")
@logtime logger pm = GroupRings.create_pm(E_2R, E_rdict, sizes[radius], twisted=true)
RG = GroupRing(G, E_2R, E_rdict, pm)
Δ = PropertyT.splaplacian(RG, S)
Δ = PropertyT.spLaplacian(RG, S)
@assert GroupRings.augmentation(Δ) == 0
save(joinpath(name, "delta.jld"), "Δ", Δ.coeffs)
save(joinpath(name, "pm.jld"), "pm", pm)
save(filename(name, :Δ), "Δ", Δ.coeffs)
save(filename(name, :pm), "pm", pm)
info(logger, "Decomposing E into orbits of $(autS)")
@logtime logger orbs = orbit_decomposition(autS, E_2R, E_rdict)
@ -195,7 +196,7 @@ function compute_orbit_data{T<:GroupElem}(logger, name::String, G::Nemo.Group, S
info(logger, "Action matrices")
@logtime logger reps = perm_reps(autS, E_2R[1:sizes[radius]], E_rdict)
save_preps(joinpath(name, "preps.jld"), reps)
save_preps(filename(name, :preps), reps)
reps = matrix_reps(reps)
info(logger, "Projections")

46
src/Projections.jl
View File

@ -175,21 +175,21 @@ function rankOne_projections(BN::WreathProduct, T::Type=Rational{Int})
append!(all_projs, [Qs[N]*RBN(p, g->BN(g)) for p in SNprojs_nc[N]])
return all_projs
end
end
##############################################################################
#
# General Groups Misc
#
##############################################################################
##############################################################################
#
# General Groups Misc
#
##############################################################################
doc"""
doc"""
products(X::Vector{GroupElem}, Y::Vector{GroupElem}, op=*)
> Returns a vector of all possible products (or `op(x,y)`), where $x\in X$ and
> $y\in Y$ are group elements. You may specify which operation is used when
> forming 'products' by adding `op` (which is `*` by default).
"""
function products{T<:GroupElem}(X::AbstractVector{T}, Y::AbstractVector{T}, op=*)
> Returns a vector of all possible products (or `op(x,y)`), where $x\in X$ and
> $y\in Y$ are group elements. You may specify which operation is used when
> forming 'products' by adding `op` (which is `*` by default).
"""
function products{T<:GroupElem}(X::AbstractVector{T}, Y::AbstractVector{T}, op=*)
result = Vector{T}()
seen = Set{T}()
for x in X
@ -202,18 +202,18 @@ function products{T<:GroupElem}(X::AbstractVector{T}, Y::AbstractVector{T}, op=*
end
end
return result
end
end
doc"""
doc"""
generateGroup(gens::Vector{GroupElem}, r=2, Id=parent(first(gens))(), op=*)
> Produces all elements of a group generated by elements in `gens` in ball of
> radius `r` (word-length metric induced by `gens`).
> If `r(=2)` is specified the procedure will terminate after generating ball
> of radius `r` in the word-length metric induced by `gens`.
> The identity element `Id` and binary operation function `op` can be supplied
> to e.g. take advantage of additive group structure.
"""
function generateGroup{T<:GroupElem}(gens::Vector{T}, r=2, Id::T=parent(first(gens))(), op=*)
> Produces all elements of a group generated by elements in `gens` in ball of
> radius `r` (word-length metric induced by `gens`).
> If `r(=2)` is specified the procedure will terminate after generating ball
> of radius `r` in the word-length metric induced by `gens`.
> The identity element `Id` and binary operation function `op` can be supplied
> to e.g. take advantage of additive group structure.
"""
function generateGroup{T<:GroupElem}(gens::Vector{T}, r=2, Id::T=parent(first(gens))(), op=*)
n = 0
R = 1
elts = gens
@ -225,4 +225,4 @@ function generateGroup{T<:GroupElem}(gens::Vector{T}, r=2, Id::T=parent(first(ge
elts = products(elts, gens, op)
end
return elts
end
end

364
src/PropertyT.jl
View File

@ -12,18 +12,28 @@ using MathProgBase
using Memento
const logger = Memento.config("info", fmt="{msg}")
const solver_logger = Memento.config("info", fmt="{msg}")
function setup_logging(name::String)
isdir(name) || mkdir(name)
L = Memento.config("info", fmt="{date}| {msg}")
Memento.add_handler(logger,
Memento.DefaultHandler(joinpath(name,"full_$(string((now()))).log"),
Memento.DefaultFormatter("{date}| {msg}")), "full_log")
handler = Memento.DefaultHandler(
filename(name, :logall), Memento.DefaultFormatter("{date}| {msg}"))
e = redirect_stderr(logger.handlers["full_log"].io)
handler.levels.x = L.levels
L.handlers["all"] = handler
# e = redirect_stderr(L.handlers["all"].io)
return L
end
function solverlogger(name)
logger = Memento.config("info", fmt="{msg}")
handler = DefaultHandler(
filename(name, :logsolver), DefaultFormatter("{date}| {msg}"))
handler.levels.x = logger.levels
logger.handlers["solver_log"] = handler
return logger
end
@ -36,7 +46,7 @@ macro logtime(logger, ex)
local diff = Base.GC_Diff(Base.gc_num(), stats)
local ts = time_string(elapsedtime, diff.allocd, diff.total_time,
Base.gc_alloc_count(diff))
esc(info(logger, ts))
$(esc(info))($(esc(logger)), ts)
val
end
end
@ -66,277 +76,112 @@ function time_string(elapsedtime, bytes, gctime, allocs)
return str
end
function exists(fname::String)
return isfile(fname) || islink(fname)
exists(fname::String) = isfile(fname) || islink(fname)
filename(prefix, s::Symbol) = filename(prefix, Val{s})
@eval begin
for (s,n) in [
[:pm, "pm.jld"],
[:Δ, "delta.jld"],
[:λ, "lambda.jld"],
[:P, "SDPmatrix.jld"],
[:warm, "warmstart.jld"],
[:Uπs, "U_pis.jld"],
[:orb, "orbits.jld"],
[:preps,"preps.jld"],
[:logall, "full_$(string(now())).log"],
[:logsolver,"solver_$(string(now())).log"]
]
filename(prefix::String, ::Type{Val{$:(s)}}) = joinpath(prefix, :($n))
end
end
function pmΔfilenames(prefix::String)
isdir(prefix) || mkdir(prefix)
pm_filename = joinpath(prefix, "pm.jld")
Δ_coeff_filename = joinpath(prefix, "delta.jld")
return pm_filename, Δ_coeff_filename
function Laplacian(name::String, G::Group)
if exists(filename(name, :Δ)) && exists(filename(name, :pm))
RG = GroupRing(G, load(filename(name, :pm), "pm"))
Δ = GroupRingElem(load(filename(name, :Δ), "Δ")[:, 1], RG)
else
throw("You need to precompute $(filename(name, :pm)) and $(filename(name, :Δ)) to load it!")
end
return Δ
end
function λSDPfilenames(prefix::String)
isdir(prefix) || mkdir(prefix)
λ_filename = joinpath(prefix, "lambda.jld")
SDP_filename = joinpath(prefix, "SDPmatrix.jld")
return λ_filename, SDP_filename
end
function Laplacian{T<:GroupElem}(S::Vector{T}, Id::T,
logger=getlogger(); radius::Int=2)
function ΔandSDPconstraints(prefix::String, G::Group)
info(logger, "Loading precomputed pm, Δ, sdp_constraints...")
pm_fname, Δ_fname = pmΔfilenames(prefix)
product_matrix = load(pm_fname, "pm")
sdp_constraints = constraints(product_matrix)
RG = GroupRing(G, product_matrix)
Δ = GroupRingElem(load(Δ_fname, "Δ")[:, 1], RG)
return Δ, sdp_constraints
end
function ΔandSDPconstraints{T<:GroupElem}(name::String, S::Vector{T}, Id::T; radius::Int=2)
info(logger, "Computing pm, Δ, sdp_constraints...")
Δ, sdp_constraints = ΔandSDPconstraints(S, Id, radius=radius)
pm_fname, Δ_fname = pmΔfilenames(name)
save(pm_fname, "pm", parent(Δ).pm)
save(Δ_fname, "Δ", Δ.coeffs)
return Δ, sdp_constraints
end
function ΔandSDPconstraints{T<:GroupElem}(S::Vector{T}, Id::T; radius::Int=2)
info(logger, "Generating balls of sizes $sizes")
info(logger, "Generating metric ball of radius $radius...")
@logtime logger E_R, sizes = Groups.generate_balls(S, Id, radius=2*radius)
info(logger, "Generated balls of sizes $sizes.")
info(logger, "Creating product matrix...")
@logtime logger pm = GroupRings.create_pm(E_R, GroupRings.reverse_dict(E_R), sizes[radius]; twisted=true)
info(logger, "Creating sdp_constratints...")
@logtime logger sdp_constraints = PropertyT.constraints(pm)
RG = GroupRing(parent(Id), E_R, pm)
Δ = splaplacian(RG, S)
return Δ, sdp_constraints
Δ = spLaplacian(RG, S)
return Δ
end
function λandP(name::String)
λ_fname, SDP_fname = λSDPfilenames(name)
f₁ = exists(λ_fname)
f₂ = exists(SDP_fname)
λ_fname = filename(name, :λ)
P_fname = filename(name, :P)
if f₁ && f₂
info(logger, "Loading precomputed λ, P...")
if exists(λ_fname) && exists(P_fname)
λ = load(λ_fname, "λ")
P = load(SDP_fname, "P")
P = load(P_fname, "P")
else
throw(ArgumentError("You need to precompute λ and SDP matrix P to load it!"))
throw("You need to precompute $λ_fname and $P_fname to load it!")
end
return λ, P
end
function λandP(name::String, SDP_problem::JuMP.Model, varλ, varP, warmstart=false)
add_handler(solver_logger,
DefaultHandler(joinpath(name, "solver_$(string(now())).log"),
DefaultFormatter("{date}| {msg}")),
"solver_log")
if warmstart && isfile(joinpath(name, "warmstart.jld"))
ws = load(joinpath(name, "warmstart.jld"), "warmstart")
function λandP(name::String, SDP::JuMP.Model, varλ, varP, warmstart=true)
if warmstart && isfile(filename(name, :warm))
ws = load(filename(name, :warm), "warmstart")
else
ws = nothing
end
λ, P, warmstart = compute_λandP(SDP_problem, varλ, varP, warmstart=ws)
solver_log = solverlogger(name)
remove_handler(solver_logger, "solver_log")
Base.Libc.flush_cstdio()
o = redirect_stdout(solver_log.handlers["solver_log"].io)
Base.Libc.flush_cstdio()
λ_fname, P_fname = λSDPfilenames(name)
λ, P, warmstart = solve_SDP(SDP, varλ, varP, warmstart=ws)
Base.Libc.flush_cstdio()
redirect_stdout(o)
delete!(solver_log.handlers, "solver_log")
if λ > 0
save(λ_fname, "λ", λ)
save(P_fname, "P", P)
save(joinpath(name, "warmstart.jld"), "warmstart", warmstart)
save(filename(name, :λ), "λ", λ)
save(filename(name, :P), "P", P)
save(filename(name, :warm), "warmstart", warmstart)
else
throw(ErrorException("Solver did not produce a valid solution!: λ = $λ"))
throw(ErrorException("Solver did not produce a valid solution: λ = $λ"))
end
return λ, P
end
function fillfrominternal!(m::JuMP.Model, traits)
# Copied from JuMP/src/solvers.jl:178
stat::Symbol = MathProgBase.status(m.internalModel)
numRows, numCols = length(m.linconstr), m.numCols
m.objBound = NaN
m.objVal = NaN
m.colVal = fill(NaN, numCols)
m.linconstrDuals = Array{Float64}(0)
discrete = (traits.int || traits.sos)
if stat == :Optimal
# If we think dual information might be available, try to get it
# If not, return an array of the correct length
if discrete
m.redCosts = fill(NaN, numCols)
m.linconstrDuals = fill(NaN, numRows)
else
if !traits.conic
m.redCosts = try
MathProgBase.getreducedcosts(m.internalModel)[1:numCols]
catch
fill(NaN, numCols)
end
m.linconstrDuals = try
MathProgBase.getconstrduals(m.internalModel)[1:numRows]
catch
fill(NaN, numRows)
end
elseif !traits.qp && !traits.qc
JuMP.fillConicDuals(m)
end
end
else
# Problem was not solved to optimality, attempt to extract useful
# information anyway
if traits.lin
if stat == :Infeasible
m.linconstrDuals = try
infray = MathProgBase.getinfeasibilityray(m.internalModel)
@assert length(infray) == numRows
infray
catch
suppress_warnings || warn("Infeasibility ray (Farkas proof) not available")
fill(NaN, numRows)
end
elseif stat == :Unbounded
m.colVal = try
unbdray = MathProgBase.getunboundedray(m.internalModel)
@assert length(unbdray) == numCols
unbdray
catch
suppress_warnings || warn("Unbounded ray not available")
fill(NaN, numCols)
end
end
end
# conic duals (currently, SOC and SDP only)
if !discrete && traits.conic && !traits.qp && !traits.qc
if stat == :Infeasible
JuMP.fillConicDuals(m)
end
end
end
# If the problem was solved, or if it terminated prematurely, try
# to extract a solution anyway. This commonly occurs when a time
# limit or tolerance is set (:UserLimit)
if !(stat == :Infeasible || stat == :Unbounded)
try
# Do a separate try since getobjval could work while getobjbound does not and vice versa
objBound = MathProgBase.getobjbound(m.internalModel) + m.obj.aff.constant
m.objBound = objBound
end
try
objVal = MathProgBase.getobjval(m.internalModel) + m.obj.aff.constant
colVal = MathProgBase.getsolution(m.internalModel)[1:numCols]
# Rescale off-diagonal terms of SDP variables
if traits.sdp
offdiagvars = JuMP.offdiagsdpvars(m)
colVal[offdiagvars] /= sqrt(2)
end
# Don't corrupt the answers if one of the above two calls fails
m.objVal = objVal
m.colVal = colVal
end
end
return stat
end
function compute_λandP(m, varλ, varP; warmstart=nothing)
λ = 0.0
P = nothing
traits = JuMP.ProblemTraits(m, relaxation=false)
while λ == 0.0
try
JuMP.build(m, traits=traits)
if warmstart != nothing
p_sol, d_sol, s = warmstart
MathProgBase.SolverInterface.setwarmstart!(m.internalModel, p_sol; dual_sol = d_sol, slack=s);
end
solve_SDP(m)
λ = MathProgBase.getobjval(m.internalModel)
catch y
warn(solver_logger, y)
end
end
warmstart = (m.internalModel.primal_sol, m.internalModel.dual_sol,
m.internalModel.slack)
fillfrominternal!(m, traits)
P = JuMP.getvalue(varP)
λ = JuMP.getvalue(varλ)
return λ, P, warmstart
end
Kazhdan_from_sgap(λ,N) = sqrt(2*λ/N)
function check_property_T(name::String, S, Id, solver, upper_bound, tol, radius)
function check_λ(name, S, λ, P, radius, logger)
isdir(name) || mkdir(name)
RG = GroupRing(parent(first(S)), load(filename(name, :pm), "pm"))
Δ = GroupRingElem(load(filename(name, :Δ), "Δ")[:, 1], RG)
if all(exists.(pmΔfilenames(name)))
# cached
Δ, sdp_constraints = ΔandSDPconstraints(name, parent(S[1]))
else
# compute
Δ, sdp_constraints = ΔandSDPconstraints(name, S, Id, radius=radius)
end
if all(exists.(λSDPfilenames(name)))
λ, P = λandP(name)
else
info(logger, "Creating SDP problem...")
SDP_problem, λ, P = create_SDP_problem(Δ, sdp_constraints, upper_bound=upper_bound)
JuMP.setsolver(SDP_problem, solver)
λ, P = λandP(name, SDP_problem, λ, P)
end
info(logger, "λ = $λ")
info(logger, "sum(P) = $(sum(P))")
info(logger, "maximum(P) = $(maximum(P))")
info(logger, "minimum(P) = $(minimum(P))")
if λ > 0
pm_fname, Δ_fname = pmΔfilenames(name)
RG = GroupRing(parent(first(S)), load(pm_fname, "pm"))
Δ = GroupRingElem(load(Δ_fname, "Δ")[:, 1], RG)
isapprox(eigvals(P), abs(eigvals(P)), atol=tol) ||
warn("The solution matrix doesn't seem to be positive definite!")
# @assert P == Symmetric(P)
@logtime logger Q = real(sqrtm(Symmetric(P)))
sgap = distance_to_positive_cone(Δ, λ, Q, 2*radius)
if isa(sgap, Interval)
sgap = sgap.lo
end
sgap = check_distance_to_cone(Δ, λ, Q, 2*radius, logger)
if sgap > 0
info(logger, "λ ≥ $(Float64(trunc(sgap,12)))")
info(logger, "λ($name, S) ≥ $(Float64(trunc(sgap,12)))")
Kazhdan_κ = Kazhdan_from_sgap(sgap, length(S))
Kazhdan_κ = Float64(trunc(Kazhdan_κ, 12))
info(logger, "κ($name, S) ≥ $Kazhdan_κ: Group HAS property (T)!")
@ -346,8 +191,51 @@ function check_property_T(name::String, S, Id, solver, upper_bound, tol, radius)
info(logger, "λ($name, S) ≥ $sgap: Group may NOT HAVE property (T)!")
return false
end
end
function check_property_T(name::String, S, Id, solver, upper_bound, tol, radius)
isdir(name) || mkdir(name)
LOGGER = Memento.getlogger()
if exists(filename(name, :pm)) && exists(filename(name, :Δ))
# cached
info(LOGGER, "Loading precomputed Δ...")
Δ = Laplacian(name, parent(S[1]))
else
# compute
Δ = Laplacian(S, Id, LOGGER, radius=radius)
save(filename(name, :pm), "pm", parent(Δ).pm)
save(filename(name, :Δ), "Δ", Δ.coeffs)
end
info(logger, "κ($name, S) ≥ $λ < 0: Tells us nothing about property (T)")
fullpath = joinpath(name, string(upper_bound))
isdir(fullpath) || mkdir(fullpath)
if exists(filename(fullpath, :λ)) && exists(filename(fullpath, :P))
info(LOGGER, "Loading precomputed λ, P...")
λ, P = λandP(fullpath)
else
info(LOGGER, "Creating SDP problem...")
SDP_problem, varλ, varP = create_SDP_problem(Δ, constraints(parent(Δ).pm), upper_bound=upper_bound)
JuMP.setsolver(SDP_problem, solver)
info(LOGGER, Base.repr(SDP_problem))
@logtime LOGGER λ, P = λandP(fullpath, SDP_problem, varλ, varP)
end
info(LOGGER, "λ = $λ")
info(LOGGER, "sum(P) = $(sum(P))")
info(LOGGER, "maximum(P) = $(maximum(P))")
info(LOGGER, "minimum(P) = $(minimum(P))")
isapprox(eigvals(P), abs.(eigvals(P)), atol=tol) ||
warn("The solution matrix doesn't seem to be positive definite!")
if λ > 0
return check_λ(name, S, λ, P, radius, LOGGER)
end
info(LOGGER, "κ($name, S) ≥ $λ < 0: Tells us nothing about property (T)")
return false
end

131
src/SDPs.jl
View File

@ -13,7 +13,7 @@ function constraints(pm, total_length=maximum(pm))
return constraints
end
function splaplacian(RG::GroupRing, S, T::Type=Float64)
function spLaplacian(RG::GroupRing, S, T::Type=Float64)
result = RG(T)
result[RG.group()] = T(length(S))
for s in S
@ -22,7 +22,7 @@ function splaplacian(RG::GroupRing, S, T::Type=Float64)
return result
end
function splaplacian{TT<:Ring}(RG::GroupRing{TT}, S, T::Type=Float64)
function spLaplacian{TT<:Ring}(RG::GroupRing{TT}, S, T::Type=Float64)
result = RG(T)
result[one(RG.group)] = T(length(S))
for s in S
@ -55,21 +55,122 @@ function create_SDP_problem(Δ::GroupRingElem, matrix_constraints; upper_bound=I
return m, λ, P
end
function solve_SDP(SDP_problem)
info(logger, Base.repr(SDP_problem))
function solve_SDP(m, varλ, varP; warmstart=nothing)
o = redirect_stdout(solver_logger.handlers["solver_log"].io)
Base.Libc.flush_cstdio()
traits = JuMP.ProblemTraits(m, relaxation=false)
@logtime logger solution_status = MathProgBase.optimize!(SDP_problem.internalModel)
Base.Libc.flush_cstdio()
redirect_stdout(o)
if solution_status != :Optimal
warn(logger, "The solver did not solve the problem successfully!")
JuMP.build(m, traits=traits)
if warmstart != nothing
p_sol, d_sol, s = warmstart
MathProgBase.SolverInterface.setwarmstart!(m.internalModel, p_sol; dual_sol = d_sol, slack=s);
end
info(logger, solution_status)
return 0
MathProgBase.optimize!(m.internalModel)
λ = MathProgBase.getobjval(m.internalModel)
warmstart = (m.internalModel.primal_sol, m.internalModel.dual_sol,
m.internalModel.slack)
fillfrominternal!(m, traits)
P = JuMP.getvalue(varP)
λ = JuMP.getvalue(varλ)
return λ, P, warmstart
end
function fillfrominternal!(m::JuMP.Model, traits)
# Copied from JuMP/src/solvers.jl:178
stat::Symbol = MathProgBase.status(m.internalModel)
numRows, numCols = length(m.linconstr), m.numCols
m.objBound = NaN
m.objVal = NaN
m.colVal = fill(NaN, numCols)
m.linconstrDuals = Array{Float64}(0)
discrete = (traits.int || traits.sos)
if stat == :Optimal
# If we think dual information might be available, try to get it
# If not, return an array of the correct length
if discrete
m.redCosts = fill(NaN, numCols)
m.linconstrDuals = fill(NaN, numRows)
else
if !traits.conic
m.redCosts = try
MathProgBase.getreducedcosts(m.internalModel)[1:numCols]
catch
fill(NaN, numCols)
end
m.linconstrDuals = try
MathProgBase.getconstrduals(m.internalModel)[1:numRows]
catch
fill(NaN, numRows)
end
elseif !traits.qp && !traits.qc
JuMP.fillConicDuals(m)
end
end
else
# Problem was not solved to optimality, attempt to extract useful
# information anyway
if traits.lin
if stat == :Infeasible
m.linconstrDuals = try
infray = MathProgBase.getinfeasibilityray(m.internalModel)
@assert length(infray) == numRows
infray
catch
suppress_warnings || warn("Infeasibility ray (Farkas proof) not available")
fill(NaN, numRows)
end
elseif stat == :Unbounded
m.colVal = try
unbdray = MathProgBase.getunboundedray(m.internalModel)
@assert length(unbdray) == numCols
unbdray
catch
suppress_warnings || warn("Unbounded ray not available")
fill(NaN, numCols)
end
end
end
# conic duals (currently, SOC and SDP only)
if !discrete && traits.conic && !traits.qp && !traits.qc
if stat == :Infeasible
JuMP.fillConicDuals(m)
end
end
end
# If the problem was solved, or if it terminated prematurely, try
# to extract a solution anyway. This commonly occurs when a time
# limit or tolerance is set (:UserLimit)
if !(stat == :Infeasible || stat == :Unbounded)
try
# Do a separate try since getobjval could work while getobjbound does not and vice versa
objBound = MathProgBase.getobjbound(m.internalModel) + m.obj.aff.constant
m.objBound = objBound
end
try
objVal = MathProgBase.getobjval(m.internalModel) + m.obj.aff.constant
colVal = MathProgBase.getsolution(m.internalModel)[1:numCols]
# Rescale off-diagonal terms of SDP variables
if traits.sdp
offdiagvars = JuMP.offdiagsdpvars(m)
colVal[offdiagvars] /= sqrt(2)
end
# Don't corrupt the answers if one of the above two calls fails
m.objVal = objVal
m.colVal = colVal
end
end
return stat
end