[R-sig-ME] Pedigreemm on lme4-1.0

[David Duffy]

On Wed, 19 Feb 2014, Ben Bolker wrote:

> The performance issue you're referring to is I think related to 
> https://github.com/lme4/lme4/issues/150 ... we're hoping to look into 
> this soon; it is definitely important, and we're afraid it's pretty 
> deep.

Using that InstEval example gives me:

Program                  Time      Notes
Wombat (AI-REML)         14.2 s

lmer4.0                  11.5 s    lme4.0_0.999999-4, Matrix_1.1-0
lmer4-nm                 25.9 s    lme4_1.0-6, Matrix_1.1-0

lmer4-bobqa              14.8 s    lme4_1.0-6, Matrix_1.1-0

lmer4-optimx-bobqa       25.2 s    lme4_1.0-6, Matrix_1.1-0
lmer4-optimx-bobqa-kkt=F 15.3 s    lme4_1.0-6, Matrix_1.1-0

lmer4-nlminb             19.8 s    lme4_1.0-6, Matrix_1.1-0
lmer4-nlminb-kkt=F       10.0 s    lme4_1.0-6, Matrix_1.1-0

lmer4-bfgs               23.4 s    lme4_1.0-6, Matrix_1.1-0

That is, the test for the Kuhn, Karush, Tucker optimality condition
seems quite expensive for this problem, and the differences between
optimizers might completely explain the 4.0 v 4 differences.

Cheers, David Duffy.

| David Duffy (MBBS PhD)
| email: David.Duffy at qimrberghofer.edu.au  ph: INT+61+7+3362-0217 fax: -0101
| Genetic Epidemiology, QIMR Berghofer Institute of Medical Research
| 300 Herston Rd, Brisbane, Queensland 4006, Australia  GPG 4D0B994A