[R-sig-ME] help: error in (function...): Downdated VtV is not positive definite and convergence problems

Discussion:

Mario Garrido

2018-10-06 14:26:23 UTC

Hello,
I tried to fit a GLMM and I get the following error. I know that my data is
probably more negative binomial than Poisson (sd>>mean), but I want to
understand where this problem comes from

glmer(countMyc.qPCR ~ sp+day +(0+day|exp.ID), family=poisson)

Error in (function (fr, X, reTrms, family, nAGQ = 1L, verbose = 0L, maxit =
100L, :
Downdated VtV is not positive definite

*countMyc.qPCR*:amount of bacteria ina particualr individual Numeric
discrete value
*sp*:species each individual vbelongs to Factor w/ 3 levels "GA","GG","GP"
*day*: day of infection Numeric discrete value
*exp.ID*: number of individual under experiment Factor w/ 33 levels
"EA1","EA10","EA12",..: 3 6 7 10 11 14 18 21 22 31 ...

I fixed the random factor as 0+day|exp.ID cause at day zero the amount of
bacteria is zero

Can be the error due the differences in scales between the minimum and
maximum value

describe.by(countMyc.qPCR)

vars n mean sd median trimmed mad min
max range skew kurtosis se
X1 1 363 127789.2 783829.6 6 455.65 8.9 0
8434322 8434322 7.84 67.75 41140.39

In addition, when I tried to fix an simpler data I have also warnings, but
other kinds

glmer(countMyc.qPCR ~day +(0+day|exp.ID), family=poisson)
Generalized linear mixed model fit by maximum likelihood (Laplace
Approximation) ['glmerMod']
Family: poisson ( log )
Formula: countMyc.qPCR ~ day + (0 + day | exp.ID)
AIC BIC logLik deviance df.resid
213021061 213021073 -106510528 213021055 360
Random effects:
Groups Name Std.Dev.
exp.ID day 0.1742
Number of obs: 363, groups: exp.ID, 33
Fixed Effects:
(Intercept) day
13.3201 -0.1898
convergence code 0; 1 optimizer warnings; 0 lme4 warnings
Warning message:
In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv, :
Model is nearly unidentifiable: very large eigenvalue
- Rescale variables?

PS. I saw a similar question before (19th July) but cannot find a solution
there
--
Mario Garrido Escudero, PhD
Dr. Hadas Hawlena Lab
Mitrani Department of Desert Ecology
Jacob Blaustein Institutes for Desert Research
Ben-Gurion University of the Negev
Midreshet Ben-Gurion 84990 ISRAEL

***@gmail.com; ***@post.bgu.ac.il
phone: (+972) 08-659-6854

[[alternative HTML version deleted]]

D. Rizopoulos

2018-10-06 16:56:34 UTC

Permalink

You could give a try to the GLMMadaptive package that can fit the same model using the adaptive Gaussian quadrature, i.e.,

library(GLMMadaptive)
fm <- mixed_model(countMyc.qPCR ~ sp + day, random = ~ 0 day | exp.ID, data = your_data, family = poisson())
summary(fm)

Best,
Dimitris

- - - - - -
Dimitris Rizopoulos
Professor of Biostatistics
Erasmus University Medical Center
The Netherlands

From: Mario Garrido <***@post.bgu.ac.il<mailto:***@post.bgu.ac.il>>
Date: Saturday, 06 Oct 2018, 4:26 PM
To: r-sig-mixed-***@r-project.org <r-sig-mixed-***@r-project.org<mailto:r-sig-mixed-***@r-project.org>>
Subject: [R-sig-ME] help: error in (function...): Downdated VtV is not positive definite and convergence problems

Hello,
I tried to fit a GLMM and I get the following error. I know that my data is
probably more negative binomial than Poisson (sd>>mean), but I want to
understand where this problem comes from

glmer(countMyc.qPCR ~ sp+day +(0+day|exp.ID), family=poisson)

Error in (function (fr, X, reTrms, family, nAGQ = 1L, verbose = 0L, maxit =
100L, :
Downdated VtV is not positive definite

*countMyc.qPCR*:amount of bacteria ina particualr individual Numeric
discrete value
*sp*:species each individual vbelongs to Factor w/ 3 levels "GA","GG","GP"
*day*: day of infection Numeric discrete value
*exp.ID*: number of individual under experiment Factor w/ 33 levels
"EA1","EA10","EA12",..: 3 6 7 10 11 14 18 21 22 31 ...

I fixed the random factor as 0+day|exp.ID cause at day zero the amount of
bacteria is zero

Can be the error due the differences in scales between the minimum and
maximum value

describe.by(countMyc.qPCR)

Mario Garrido

2018-10-08 11:09:12 UTC

Permalink

Thanks so much Dimitris,
in any case, now I have recieved another warning

fm <- mixed_model(countMyc.qPCR ~ sp + day, random = ~ 0+day | exp.ID, data
= MycGLMM, family = poisson())
Error in optim(par = b_i, fn = log_post_b, gr = score_log_post_b, method =
"BFGS", :
non-finite value supplied by optim

Any idea why?

Thanks!

PS, sorry to insist, but is not a problem what I said in the mail
before? due the differences in scales between the minimum and maximum value

Post by D. Rizopoulos
You could give a try to the GLMMadaptive package that can fit the same
model using the adaptive Gaussian quadrature, i.e.,
library(GLMMadaptive)
fm <- mixed_model(countMyc.qPCR ~ sp + day, random = ~ 0 day | exp.ID,
data = your_data, family = poisson())
summary(fm)
Best,
Dimitris
- - - - - -
Dimitris Rizopoulos
Professor of Biostatistics
Erasmus University Medical Center
The Netherlands
*Date: *Saturday, 06 Oct 2018, 4:26 PM
*Subject: *[R-sig-ME] help: error in (function...): Downdated VtV is not
positive definite and convergence problems
Hello,
I tried to fit a GLMM and I get the following error. I know that my data is
probably more negative binomial than Poisson (sd>>mean), but I want to
understand where this problem comes from
glmer(countMyc.qPCR ~ sp+day +(0+day|exp.ID), family=poisson)
Error in (function (fr, X, reTrms, family, nAGQ = 1L, verbose = 0L, maxit =
Downdated VtV is not positive definite
*countMyc.qPCR*:amount of bacteria ina particualr individual Numeric
discrete value
*sp*:species each individual vbelongs to Factor w/ 3 levels "GA","GG","GP"
*day*: day of infection Numeric discrete value
*exp.ID*: number of individual under experiment Factor w/ 33 levels
"EA1","EA10","EA12",..: 3 6 7 10 11 14 18 21 22 31 ...
I fixed the random factor as 0+day|exp.ID cause at day zero the amount of
bacteria is zero
Can be the error due the differences in scales between the minimum and
maximum value

describe.by(countMyc.qPCR)

vars n mean sd median trimmed mad min
max range skew kurtosis se
X1 1 363 127789.2 783829.6 6 455.65 8.9 0
8434322 8434322 7.84 67.75 41140.39
In addition, when I tried to fix an simpler data I have also warnings, but
other kinds
glmer(countMyc.qPCR ~day +(0+day|exp.ID), family=poisson)
Generalized linear mixed model fit by maximum likelihood (Laplace
Approximation) ['glmerMod']
Family: poisson ( log )
Formula: countMyc.qPCR ~ day + (0 + day | exp.ID)
AIC BIC logLik deviance df.resid
213021061 213021073 -106510528 213021055 360
Groups Name Std.Dev.
exp.ID day 0.1742
Number of obs: 363, groups: exp.ID, 33
(Intercept) day
13.3201 -0.1898
convergence code 0; 1 optimizer warnings; 0 lme4 warnings
Model is nearly unidentifiable: very large eigenvalue
- Rescale variables?
PS. I saw a similar question before (19th July) but cannot find a solution
there
--
Mario Garrido Escudero, PhD
Dr. Hadas Hawlena Lab
Mitrani Department of Desert Ecology
Jacob Blaustein Institutes for Desert Research
Ben-Gurion University of the Negev
Midreshet Ben-Gurion 84990 ISRAEL
phone: (+972) 08-659-6854
[[alternative HTML version deleted]]
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

--
Mario Garrido Escudero, PhD
Dr. Hadas Hawlena Lab
Mitrani Department of Desert Ecology
Jacob Blaustein Institutes for Desert Research
Ben-Gurion University of the Negev
Midreshet Ben-Gurion 84990 ISRAEL

***@gmail.com; ***@post.bgu.ac.il
phone: (+972) 08-659-6854

[[alternative HTML version deleted]]

Mario Garrido

2018-10-08 11:10:56 UTC

Permalink

Thanks again Dr. Rizopoulos,
i cannot find the command initial values in the Package 'lme4' PDF here:
https://cran.r-project.org/web/packages/lme4/lme4.pdf

Where can I look how to use it?

thanks!

Post by Mario Garrido
Thanks so much Dimitris,
in any case, now I have recieved another warning
fm <- mixed_model(countMyc.qPCR ~ sp + day, random = ~ 0+day | exp.ID,
data = MycGLMM, family = poisson())
Error in optim(par = b_i, fn = log_post_b, gr = score_log_post_b, method =
non-finite value supplied by optim
Any idea why?
Thanks!
PS, sorry to insist, but is not a problem what I said in the mail
before? due the differences in scales between the minimum and maximum value

describe.by(countMyc.qPCR)

vars n mean sd median trimmed mad min
max range skew kurtosis se
X1 1 363 127789.2 783829.6 6 455.65 8.9 0
8434322 8434322 7.84 67.75 41140.39
In addition, when I tried to fix an simpler data I have also warnings, but
other kinds
glmer(countMyc.qPCR ~day +(0+day|exp.ID), family=poisson)
Generalized linear mixed model fit by maximum likelihood (Laplace
Approximation) ['glmerMod']
Family: poisson ( log )
Formula: countMyc.qPCR ~ day + (0 + day | exp.ID)
AIC BIC logLik deviance df.resid
213021061 213021073 -106510528 213021055 360
Groups Name Std.Dev.
exp.ID day 0.1742
Number of obs: 363, groups: exp.ID, 33
(Intercept) day
13.3201 -0.1898
convergence code 0; 1 optimizer warnings; 0 lme4 warnings
Model is nearly unidentifiable: very large eigenvalue
- Rescale variables?
PS. I saw a similar question before (19th July) but cannot find a solution
there
--
Mario Garrido Escudero, PhD
Dr. Hadas Hawlena Lab
Mitrani Department of Desert Ecology
Jacob Blaustein Institutes for Desert Research
Ben-Gurion University of the Negev
Midreshet Ben-Gurion 84990 ISRAEL
phone: (+972) 08-659-6854
[[alternative HTML version deleted]]
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

D. Rizopoulos

2018-10-08 14:56:23 UTC

Permalink

You can set the initial values in the mixed_model() function of my
GLMMadaptive package using the initial_values argument. For more info,
check: https://drizopoulos.github.io/GLMMadaptive/reference/mixed_model.html

Best,
Dimitris

Post by Mario Garrido
Thanks again Dr. Rizopoulos,
https://cran.r-project.org/web/packages/lme4/lme4.pdf
Where can I look how to use it?
thanks!

describe.by(countMyc.qPCR)

vars n mean sd median trimmed mad min
max range skew kurtosis se
X1 1 363 127789.2 783829.6 6 455.65 8.9 0
8434322 8434322 7.84 67.75 41140.39
In addition, when I tried to fix an simpler data I have also warnings, but
other kinds
glmer(countMyc.qPCR ~day +(0+day|exp.ID), family=poisson)
Generalized linear mixed model fit by maximum likelihood (Laplace
Approximation) ['glmerMod']
Family: poisson ( log )
Formula: countMyc.qPCR ~ day + (0 + day | exp.ID)
AIC BIC logLik deviance df.resid
213021061 213021073 -106510528 213021055 360
Groups Name Std.Dev.
exp.ID day 0.1742
Number of obs: 363, groups: exp.ID, 33
(Intercept) day
13.3201 -0.1898
convergence code 0; 1 optimizer warnings; 0 lme4 warnings
Model is nearly unidentifiable: very large eigenvalue
- Rescale variables?
PS. I saw a similar question before (19th July) but cannot find a solution
there
--
Mario Garrido Escudero, PhD
Dr. Hadas Hawlena Lab
Mitrani Department of Desert Ecology
Jacob Blaustein Institutes for Desert Research
Ben-Gurion University of the Negev
Midreshet Ben-Gurion 84990 ISRAEL
phone: (+972) 08-659-6854
[[alternative HTML version deleted]]
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

--
Dimitris Rizopoulos
Professor of Biostatistics
Department of Biostatistics
Erasmus University Medical Center

Address: PO Box 2040, 3000 CA Rotterdam, the Netherlands
Tel: +31/(0)10/7043478
Fax: +31/(0)10/7043014
Web (personal): http://www.drizopoulos.com/
Web (work): http://www.erasmusmc.nl/biostatistiek/
Blog: http://iprogn.blogspot.nl/