Specify maximum number of iterations

[venkatasg]

I'm new to Julia and the MixedModels package, so I apologize if this has already been implemented. Is it possible to specify the maximum number of iterations when running a model fit, in addition to the fast=true and nAGQ parameters? I find that even with fast=true, the model sometimes takes far too long for my liking. Interrupting the model fit with Ctrl+C lets me inspect model parameters earlier (in a notebook), and I find that sometimes the estimates are reasonably good enough - not too far from the final fit.

[palday]

You can modify the optsum property of the MixedModel before fitting.

using MixedModels
julia> verbagg = MixedModels.dataset(:verbagg)
Arrow.Table with 7584 rows, 9 columns, and schema:
 :subj    String
 :item    String
 :anger   Int8
 :gender  String
 :btype   String
 :situ    String
 :mode    String
 :resp    String
 :r2      String

julia> const vaform = @formula(r2 ~ 1 + anger + gender + btype + situ + (1|subj) + (1|item));

julia> mdl = GeneralizedLinearMixedModel(vaform, verbagg, Bernoulli());

julia> mdl.optsum
Initial parameter vector: [1.0, 1.0]
Initial objective value:  Inf

Optimizer (from NLopt):   LN_BOBYQA
Lower bounds:             [0.0, 0.0]
ftol_rel:                 1.0e-12
ftol_abs:                 1.0e-8
xtol_rel:                 0.0
xtol_abs:                 [1.0e-10, 1.0e-10]
initial_step:             Float64[]
maxfeval:                 -1
maxtime:                  -1.0

Function evaluations:     -1
Final parameter vector:   [1.0, 1.0]
Final objective value:    Inf
Return code:              FAILURE


julia> mdl.optsum.maxfeval = 10 # limit to 10 evalations
10

julia> fit!(mdl)
Minimizing 11    Time: 0:00:00 (12.65 ms/it)
┌ Warning: NLopt optimization failure: MAXEVAL_REACHED
└ @ MixedModels ~/.julia/packages/MixedModels/8zfgx/src/optsummary.jl:182
Generalized Linear Mixed Model fit by maximum likelihood (nAGQ = 1)
  r2 ~ 1 + anger + gender + btype + situ + (1 | subj) + (1 | item)
  Distribution: Bernoulli{Float64}
  Link: LogitLink()

   logLik    deviance     AIC       AICc        BIC    
 -4096.7549  8193.5098  8209.5098  8209.5288  8264.9802

Variance components:
        Column   VarianceStd.Dev.
subj (Intercept)  3.06 1.75
item (Intercept)  1.00 1.00

 Number of obs: 7584; levels of grouping factors: 316, 24

Fixed-effects parameters:
─────────────────────────────────────────────────────
                   Coef.  Std. Error      z  Pr(>|z|)
─────────────────────────────────────────────────────
(Intercept)    0.206053    0.604625    0.34    0.7333
anger          0.0399404   0.0213439   1.87    0.0613
gender: M      0.231317    0.243805    0.95    0.3427
btype: scold  -0.794186    0.504748   -1.57    0.1156
btype: shout  -1.53919     0.505766   -3.04    0.0023
situ: self    -0.776656    0.412493   -1.88    0.0597
─────────────────────────────────────────────────────

julia> mdl.optsum.maxfeval = -1 # unlimited evaluations
-1

julia> mdl.optsum.ftol_rel = 1e-8 # loosen convergence criterion
1.0e-8

julia> fit!(mdl)
ERROR: ArgumentError: This model has already been fitted. Use refit!() instead.
Stacktrace:
 [1] fit!(m::GeneralizedLinearMixedModel{Float64, Bernoulli{Float64}}; verbose::Bool, fast::Bool, nAGQ::Int64, progress::Bool, thin::Int64, init_from_lmm::Set{Any})
   @ MixedModels ~/.julia/packages/MixedModels/8zfgx/src/generalizedlinearmixedmodel.jl:259
 [2] fit!(m::GeneralizedLinearMixedModel{Float64, Bernoulli{Float64}})
   @ MixedModels ~/.julia/packages/MixedModels/8zfgx/src/generalizedlinearmixedmodel.jl:241
 [3] top-level scope
   @ REPL[27]:1

julia> refit!(mdl)
Minimizing 149   Time: 0:00:00 ( 3.26 ms/it)
Generalized Linear Mixed Model fit by maximum likelihood (nAGQ = 1)
  r2 ~ 1 + anger + gender + btype + situ + (1 | subj) + (1 | item)
  Distribution: Bernoulli{Float64}
  Link: LogitLink()

   logLik    deviance     AIC       AICc        BIC    
 -4075.7308  8151.4615  8167.4615  8167.4806  8222.9319

Variance components:
        Column   Variance Std.Dev. 
subj (Intercept)  1.794784 1.339696
item (Intercept)  0.247374 0.497367

 Number of obs: 7584; levels of grouping factors: 316, 24

Fixed-effects parameters:
─────────────────────────────────────────────────────
                   Coef.  Std. Error      z  Pr(>|z|)
─────────────────────────────────────────────────────
(Intercept)    0.295769    0.405585    0.73    0.4659
anger          0.0545254   0.0167561   3.25    0.0011
gender: M      0.308611    0.191255    1.61    0.1066
btype: scold  -1.07484     0.257802   -4.17    <1e-04
btype: shout  -2.12194     0.259521   -8.18    <1e-15
situ: self    -1.07715     0.211115   -5.10    <1e-06
─────────────────────────────────────────────────────