[Gretl-users] Re: "Verbose" option in mle command
On Wed, 25 Mar 2020, Alecos Papadopoulos wrote:
At least as best as I can remember, also in older versions of gretl
(for
windows 64bit), the "verbose" option in estimation commands like gmm and mle
appeared to be a bit of a loose agent, not producing always the same info:
sometimes all the iterations were shown, sometimes one every five iterations
was shown... things like that.
But in the current version gretl 2020a, the mle --verbose command gives
consistently a compressed picture of each iteration like
1: loglikelihood -1363.99055710 (norm 1.76e+001)
2: loglikelihood -1278.49934122 (step 0.00032, norm 1.21e+001)
3: loglikelihood -1239.79066487 (step 0.00032, norm 1.02e+001)
4: loglikelihood -1237.94568712 (step 0.0016, norm 1.09e+001)
5: loglikelihood -1235.14819783 (step 0.008, norm 1.08e+001)
etc
Gone are the coefficient values per iteration, gone are the gradient values
per iteration... If indeed this is the case, can I ask why? I had found them
to be really helpful when I was dealing with convergence issues.
That was my idea, the previous behaviour made things hard to follow when
you have large models, with dozens of parameters and you just want to
monitor what's going on.
You can reinstate the old output via the "set" variable "max_verbose",
as
in
<hansl>
set seed 909090
nulldata 100
x = uniform() < 0.5
p = 0.5
set max_verbose full
mle llik = x ? log(p) : log(1-p)
params p
end mle --verbose
</hansl>
The helpfile says:
<help>
max_verbose: off (the default), on or full. Controls the verbosity of
commands and functions that use numerical optimization methods. The on
choice applies only to functions (such as "BFGSmax" and "NRmax")
which
work silently by default; the effect is to print basic iteration
information. The full setting can be used to trigger more detailed output,
including parameter values and their respective gradient for the objective
function at each iteration. This choice applies both to functions of the
above-mentioned sort and to commands that rely on numerical optimization
such as "arima", "probit" and "mle". In the case of commands
the effect is
to make their --verbose option produce more detail. See also chapter 35 of
the Gretl User's Guide.
</help>
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------