Skip to content

Add test for the linear API #47

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
May 2, 2022
Merged

Add test for the linear API #47

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
00a7f8e
add consistency checks for linear API
tmigot Apr 11, 2022
e497e9a
update NLPModelsModifiers to v0.6
tmigot Apr 29, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions Project.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,8 +10,8 @@ SparseArrays = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"
Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"

[compat]
NLPModels = "0.16, 0.17, 0.18"
NLPModelsModifiers = "0.3, 0.4, 0.5"
NLPModels = "0.18"
NLPModelsModifiers = "0.6"
julia = "^1.3.0"

[extras]
Expand Down
247 changes: 244 additions & 3 deletions src/nlp/consistency.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,7 +24,7 @@ function consistent_nlps(
)
consistent_counters(nlps, linear_api = linear_api)
test_meta && consistent_meta(nlps, rtol = rtol)
consistent_functions(nlps, rtol = rtol, exclude = exclude)
consistent_functions(nlps, linear_api = linear_api, rtol = rtol, exclude = exclude)
consistent_counters(nlps, linear_api = linear_api)
for nlp in nlps
reset!(nlp)
Expand All @@ -47,6 +47,7 @@ function consistent_nlps(
]
consistent_functions(
[nlps; qnmodels],
linear_api = linear_api,
exclude = [hess, hess_coord, hprod, jth_hess, jth_hess_coord, jth_hprod, ghjvprod] ∪ exclude,
)
consistent_counters([nlps; qnmodels], linear_api = linear_api)
Expand All @@ -55,7 +56,11 @@ function consistent_nlps(
if test_slack && has_inequalities(nlps[1])
reset!.(nlps)
slack_nlps = SlackModel.(nlps)
consistent_functions(slack_nlps, exclude = [jth_hess, jth_hess_coord, jth_hprod] ∪ exclude)
consistent_functions(
slack_nlps,
linear_api = linear_api,
exclude = [jth_hess, jth_hess_coord, jth_hprod] ∪ exclude,
)
consistent_counters(slack_nlps, linear_api = linear_api)
end
end
Expand Down Expand Up @@ -103,13 +108,19 @@ function consistent_counters(nlps; linear_api = false)
end
end

function consistent_functions(nlps; rtol = 1.0e-8, exclude = [])
function consistent_functions(nlps; linear_api = false, rtol = 1.0e-8, exclude = [])
N = length(nlps)
n = nlps[1].meta.nvar
m = nlps[1].meta.ncon
test_lin = linear_api && (nlps[1].meta.nlin > 0)
mlin = nlps[1].meta.nlin
test_nln = linear_api && (nlps[1].meta.nnln > 0)
mnln = nlps[1].meta.nnln

tmp_n = zeros(n)
tmp_m = zeros(m)
tmp_mlin = zeros(mlin)
tmp_mnln = zeros(mnln)
tmp_nn = zeros(n, n)

x = 10 * [-(-1.0)^i for i = 1:n]
Expand Down Expand Up @@ -292,6 +303,68 @@ function consistent_functions(nlps; rtol = 1.0e-8, exclude = [])
end
end

if !(cons in exclude) && test_lin
cs = Any[cons_lin(nlp, x) for nlp in nlps]
cls = [nlp.meta.lcon[nlp.meta.lin] for nlp in nlps]
cus = [nlp.meta.ucon[nlp.meta.lin] for nlp in nlps]
cmin = minimum(map(norm, cs))
for i = 1:N
tmpc = cons_lin!(nlps[i], x, tmp_mlin)
@test isapprox(cs[i], tmp_mlin, atol = rtol * max(cmin, 1.0))
@test isapprox(tmpc, tmp_mlin, atol = rtol * max(cmin, 1.0))
ci, li, ui = copy(cs[i]), cls[i], cus[i]
for k = 1:mlin
if li[k] > -Inf
ci[k] -= li[k]
elseif ui[k] < Inf
ci[k] -= ui[k]
end
end
for j = (i + 1):N
cj, lj, uj = copy(cs[j]), cls[j], cus[j]
for k = 1:mlin
if lj[k] > -Inf
cj[k] -= lj[k]
elseif uj[k] < Inf
cj[k] -= uj[k]
end
end
@test isapprox(norm(ci), norm(cj), atol = rtol * max(cmin, 1.0))
end
end
end

if !(cons in exclude) && test_nln
cs = Any[cons_nln(nlp, x) for nlp in nlps]
cls = [nlp.meta.lcon[nlp.meta.nln] for nlp in nlps]
cus = [nlp.meta.ucon[nlp.meta.nln] for nlp in nlps]
cmin = minimum(map(norm, cs))
for i = 1:N
tmpc = cons_nln!(nlps[i], x, tmp_mnln)
@test isapprox(cs[i], tmp_mnln, atol = rtol * max(cmin, 1.0))
@test isapprox(tmpc, tmp_mnln, atol = rtol * max(cmin, 1.0))
ci, li, ui = copy(cs[i]), cls[i], cus[i]
for k = 1:mnln
if li[k] > -Inf
ci[k] -= li[k]
elseif ui[k] < Inf
ci[k] -= ui[k]
end
end
for j = (i + 1):N
cj, lj, uj = copy(cs[j]), cls[j], cus[j]
for k = 1:mnln
if lj[k] > -Inf
cj[k] -= lj[k]
elseif uj[k] < Inf
cj[k] -= uj[k]
end
end
@test isapprox(norm(ci), norm(cj), atol = rtol * max(cmin, 1.0))
end
end
end

if intersect([jac, jac_coord], exclude) == []
Js = [jac(nlp, x) for nlp in nlps]
Jmin = minimum(map(norm, Js))
Expand All @@ -314,6 +387,50 @@ function consistent_functions(nlps; rtol = 1.0e-8, exclude = [])
end
end

if (intersect([jac, jac_coord], exclude) == []) && test_lin
Js = [jac_lin(nlp, x) for nlp in nlps]
Jmin = minimum(map(norm, Js))
for i = 1:N
vi = norm(Js[i])
for j = (i + 1):N
@test isapprox(vi, norm(Js[j]), atol = rtol * max(Jmin, 1.0))
end
V = jac_lin_coord(nlps[i], x)
I, J = jac_lin_structure(nlps[i])
@test length(I) == length(J) == length(V) == nlps[i].meta.lin_nnzj
@test isapprox(sparse(I, J, V, mlin, n), Js[i], atol = rtol * max(Jmin, 1.0))
IS, JS = zeros(Int, nlps[i].meta.lin_nnzj), zeros(Int, nlps[i].meta.lin_nnzj)
jac_lin_structure!(nlps[i], IS, JS)
@test IS == I
@test JS == J
tmp_V = zeros(nlps[i].meta.lin_nnzj)
jac_lin_coord!(nlps[i], x, tmp_V)
@test tmp_V == V
end
end

if (intersect([jac, jac_coord], exclude) == []) && test_nln
Js = [jac_nln(nlp, x) for nlp in nlps]
Jmin = minimum(map(norm, Js))
for i = 1:N
vi = norm(Js[i])
for j = (i + 1):N
@test isapprox(vi, norm(Js[j]), atol = rtol * max(Jmin, 1.0))
end
V = jac_nln_coord(nlps[i], x)
I, J = jac_nln_structure(nlps[i])
@test length(I) == length(J) == length(V) == nlps[i].meta.nln_nnzj
@test isapprox(sparse(I, J, V, mnln, n), Js[i], atol = rtol * max(Jmin, 1.0))
IS, JS = zeros(Int, nlps[i].meta.nln_nnzj), zeros(Int, nlps[i].meta.nln_nnzj)
jac_nln_structure!(nlps[i], IS, JS)
@test IS == I
@test JS == J
tmp_V = zeros(nlps[i].meta.nln_nnzj)
jac_nln_coord!(nlps[i], x, tmp_V)
@test tmp_V == V
end
end

if !(jprod in exclude)
Jops = Any[jac_op(nlp, x) for nlp in nlps]
Jps = Any[jprod(nlp, x, v) for nlp in nlps]
Expand Down Expand Up @@ -345,6 +462,68 @@ function consistent_functions(nlps; rtol = 1.0e-8, exclude = [])
end
end

if (!(jprod in exclude)) && test_lin
Jlinops = Any[jac_lin_op(nlp, x) for nlp in nlps]
Jps = Any[jprod_lin(nlp, x, v) for nlp in nlps]
for i = 1:N
@test isapprox(Jps[i], Jlinops[i] * v, atol = rtol * max(Jmin, 1.0))
vi = norm(Jps[i])
for j = (i + 1):N
@test isapprox(vi, norm(Jps[j]), atol = rtol * max(Jmin, 1.0))
end
tmpjv = jprod_lin!(nlps[i], x, v, tmp_mlin)
@test isapprox(tmpjv, tmp_mlin, atol = rtol * max(Jmin, 1.0))
@test isapprox(Jps[i], tmp_mlin, atol = rtol * max(Jmin, 1.0))
fill!(tmp_mlin, 0)
J = jac_lin_op!(nlps[i], x, tmp_mlin, tmp_n)
res = J * v
@test isapprox(res, Jps[i], atol = rtol * max(Jmin, 1.0))
@test isapprox(res, tmp_mlin, atol = rtol * max(Jmin, 1.0))

if !(jac_coord in exclude)
rows, cols = jac_lin_structure(nlps[i])
vals = jac_lin_coord(nlps[i], x)
jprod_lin!(nlps[i], rows, cols, vals, v, tmp_mlin)
@test isapprox(Jps[i], tmp_mlin, atol = rtol * max(Jmin, 1.0))

J = jac_lin_op!(nlps[i], rows, cols, vals, tmp_mlin, tmp_n)
res = J * v
@test isapprox(res, Jps[i], atol = rtol * max(Jmin, 1.0))
end
end
end

if (!(jprod in exclude)) && test_nln
Jnlnops = Any[jac_nln_op(nlp, x) for nlp in nlps]
Jps = Any[jprod_nln(nlp, x, v) for nlp in nlps]
for i = 1:N
@test isapprox(Jps[i], Jnlnops[i] * v, atol = rtol * max(Jmin, 1.0))
vi = norm(Jps[i])
for j = (i + 1):N
@test isapprox(vi, norm(Jps[j]), atol = rtol * max(Jmin, 1.0))
end
tmpjv = jprod_nln!(nlps[i], x, v, tmp_mnln)
@test isapprox(tmpjv, tmp_mnln, atol = rtol * max(Jmin, 1.0))
@test isapprox(Jps[i], tmp_mnln, atol = rtol * max(Jmin, 1.0))
fill!(tmp_mnln, 0)
J = jac_nln_op!(nlps[i], x, tmp_mnln, tmp_n)
res = J * v
@test isapprox(res, Jps[i], atol = rtol * max(Jmin, 1.0))
@test isapprox(res, tmp_mnln, atol = rtol * max(Jmin, 1.0))

if !(jac_coord in exclude)
rows, cols = jac_nln_structure(nlps[i])
vals = jac_nln_coord(nlps[i], x)
jprod_nln!(nlps[i], rows, cols, vals, v, tmp_mnln)
@test isapprox(Jps[i], tmp_mnln, atol = rtol * max(Jmin, 1.0))

J = jac_nln_op!(nlps[i], rows, cols, vals, tmp_mnln, tmp_n)
res = J * v
@test isapprox(res, Jps[i], atol = rtol * max(Jmin, 1.0))
end
end
end

if !(jtprod in exclude)
w = 10 * [-(-1.0)^i for i = 1:m]
Jtps = Any[jtprod(nlp, x, w) for nlp in nlps]
Expand Down Expand Up @@ -376,6 +555,68 @@ function consistent_functions(nlps; rtol = 1.0e-8, exclude = [])
end
end

if (!(jtprod in exclude)) && test_lin
w = 10 * [-(-1.0)^i for i = 1:mlin]
Jtps = Any[jtprod_lin(nlp, x, w) for nlp in nlps]
for i = 1:N
@test isapprox(Jtps[i], Jlinops[i]' * w, atol = rtol * max(Jmin, 1.0))
vi = norm(Jtps[i])
for j = (i + 1):N
@test isapprox(vi, norm(Jtps[j]), atol = rtol * max(Jmin, 1.0))
end
tmpjtv = jtprod_lin!(nlps[i], x, w, tmp_n)
@test isapprox(Jtps[i], tmp_n, atol = rtol * max(Jmin, 1.0))
@test isapprox(tmpjtv, tmp_n, atol = rtol * max(Jmin, 1.0))
fill!(tmp_n, 0)
J = jac_lin_op!(nlps[i], x, tmp_mlin, tmp_n)
res = J' * w
@test isapprox(res, Jtps[i], atol = rtol * max(Jmin, 1.0))
@test isapprox(res, tmp_n, atol = rtol * max(Jmin, 1.0))

if !(jac_coord in exclude)
rows, cols = jac_lin_structure(nlps[i])
vals = jac_lin_coord(nlps[i], x)
jtprod_lin!(nlps[i], rows, cols, vals, w, tmp_n)
@test isapprox(Jtps[i], tmp_n, atol = rtol * max(Jmin, 1.0))

J = jac_lin_op!(nlps[i], rows, cols, vals, tmp_mlin, tmp_n)
res = J' * w
@test isapprox(res, Jtps[i], atol = rtol * max(Jmin, 1.0))
end
end
end

if (!(jtprod in exclude)) && test_nln
w = 10 * [-(-1.0)^i for i = 1:mnln]
Jtps = Any[jtprod_nln(nlp, x, w) for nlp in nlps]
for i = 1:N
@test isapprox(Jtps[i], Jnlnops[i]' * w, atol = rtol * max(Jmin, 1.0))
vi = norm(Jtps[i])
for j = (i + 1):N
@test isapprox(vi, norm(Jtps[j]), atol = rtol * max(Jmin, 1.0))
end
tmpjtv = jtprod_nln!(nlps[i], x, w, tmp_n)
@test isapprox(Jtps[i], tmp_n, atol = rtol * max(Jmin, 1.0))
@test isapprox(tmpjtv, tmp_n, atol = rtol * max(Jmin, 1.0))
fill!(tmp_n, 0)
J = jac_nln_op!(nlps[i], x, tmp_mnln, tmp_n)
res = J' * w
@test isapprox(res, Jtps[i], atol = rtol * max(Jmin, 1.0))
@test isapprox(res, tmp_n, atol = rtol * max(Jmin, 1.0))

if !(jac_coord in exclude)
rows, cols = jac_nln_structure(nlps[i])
vals = jac_nln_coord(nlps[i], x)
jtprod_nln!(nlps[i], rows, cols, vals, w, tmp_n)
@test isapprox(Jtps[i], tmp_n, atol = rtol * max(Jmin, 1.0))

J = jac_nln_op!(nlps[i], rows, cols, vals, tmp_mnln, tmp_n)
res = J' * w
@test isapprox(res, Jtps[i], atol = rtol * max(Jmin, 1.0))
end
end
end

y = 3.14 * ones(m)

if !(hess_coord in exclude)
Expand Down
8 changes: 4 additions & 4 deletions src/nls/consistency.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,24 +29,24 @@ function consistent_nlss(
for nls in nlss
reset!(nls)
end
consistent_functions(nlss, exclude = exclude)
consistent_functions(nlss, linear_api = linear_api, exclude = exclude)

if test_slack && has_inequalities(nlss[1])
reset!.(nlss)
slack_nlss = SlackNLSModel.(nlss)
consistent_nls_functions(slack_nlss, exclude = exclude)
consistent_nls_counters(slack_nlss)
consistent_counters(slack_nlss, linear_api = linear_api)
consistent_functions(slack_nlss, exclude = [jth_hess, jth_hess_coord, jth_hprod] ∪ exclude)
consistent_functions(slack_nlss, linear_api = linear_api, exclude = [jth_hess, jth_hess_coord, jth_hprod] ∪ exclude)
end

if test_ff
reset!.(nlss)
ff_nlss = FeasibilityFormNLS.(nlss)
consistent_nls_functions(ff_nlss, exclude = exclude)
consistent_nls_counters(ff_nlss)
consistent_counters(ff_nlss, linear_api = linear_api)
consistent_functions(ff_nlss, exclude = [jth_hess, jth_hess_coord, jth_hprod] ∪ exclude)
consistent_counters(ff_nlss, linear_api = false)
consistent_functions(ff_nlss, linear_api = false, exclude = [jth_hess, jth_hess_coord, jth_hprod] ∪ exclude)
end
end

Expand Down