Add flag to hide warnings

[nantonel]

Should be easy to implement?

For example in PANOC

if warnings
  @warn "parameter `gamma` became too small ($(state.gamma)), stopping the iterations"
end

I would also default it to warnings=false.

[lostella]

I agree in principle, but could you attach a MWE that triggers the warning? Because while hiding the warning we should also (i) make sure that no bug is causing the line search to fail (ii) if no bug is there, make the algorithm's stopping criterion more robust (or line search one, or both).

[nantonel]

Stumbled upon this MWE

using StructuredOptimization
using Random
using ProximalAlgorithms: PANOC 
using ProximalOperators, AbstractOperators
using RecursiveArrayTools

Random.seed!(13)
n,m = 10,20
Y = randn(n,m)

A = [Eye(size(Y)) Eye(size(Y))]
x0 = ArrayPartition(zeros(size(Y)), zeros(size(Y)))

f = PrecomposeDiagonal(SqrNormL2(), 1.0, Y)
g1 = NormL1() 
g2 = IndBallRank(2) 
g = SeparableSum(g2,g1) 

solver = PANOC(verbose=true)
solver(x0; f=f, A=A, g=g)

maybe there's something simpler but better then nothing for the moment!

[lostella]

On Julia 1.1.1 I get an even stranger behaviour:

[...]
  120 | 4.750e-01 | 8.124e-08 | 1.000e+00
  130 | 4.750e-01 | 1.877e-07 | 5.000e-01
┌ Warning: stepsize `tau` became too small (0.0009765625), stopping the iterations
└ @ ProximalAlgorithms ~/.julia/dev/ProximalAlgorithms/src/algorithms/panoc.jl:182
  139 | 4.750e-01 | 7.043e-07 | 7.813e-03
  149 | 4.750e-01 | 1.803e-07 | 5.000e-01
[...]

i.e., the warning is actually a lie! (The iteration continues somehow) Looking into it.

[nantonel]

Yes, I forgot to mention that I'm in Julia 1.1.1!

[lostella]

Seems like a weird bug (and a hard one to reproduce): changing the last two lines in the MWE to

solver = PANOC(verbose=true, freq=1)
solver(x0; f=f, A=A, g=g)

then the iterations do stop as expected:

[...]
  135 | 4.750e-01 | 4.292e-08 | 1.000e+00
  136 | 4.750e-01 | 4.575e-08 | 1.000e+00
  137 | 4.750e-01 | 1.417e-08 | 1.250e-01
  138 | 4.750e-01 | 2.124e-07 | 1.250e-01
  139 | 4.750e-01 | 2.189e-07 | 7.813e-03
┌ Warning: stepsize `tau` became too small (0.0009765625), stopping the iterations
└ @ ProximalAlgorithms ~/.julia/dev/ProximalAlgorithms/src/algorithms/panoc.jl:182
julia> 

[nantonel]

Very strange!! We should check in Julia 1.2.. but I'm not going to update yet (waiting next week)

[nantonel]

Seems in the latest version the MWE below doesn't raise any warnings!
At least when using PANOC, ZeroFPR is still having some.

Stumbled upon this MWE

using StructuredOptimization
using Random
using ProximalAlgorithms: PANOC 
using ProximalOperators, AbstractOperators
using RecursiveArrayTools

Random.seed!(13)
n,m = 10,20
Y = randn(n,m)

A = [Eye(size(Y)) Eye(size(Y))]
x0 = ArrayPartition(zeros(size(Y)), zeros(size(Y)))

f = PrecomposeDiagonal(SqrNormL2(), 1.0, Y)
g1 = NormL1() 
g2 = IndBallRank(2) 
g = SeparableSum(g2,g1) 

solver = PANOC(verbose=true)
solver(x0; f=f, A=A, g=g)

maybe there's something simpler but better then nothing for the moment!

[lostella]

I’m closing this, since using @warn or other builtin logging utilities already allows tuning the logging level, see e.g. disable_logging.