代码之家 › 专栏 › 技术社区 › Scransom

抑制R中的JAGS“值超出范围”警告

rjags r2jags jags r

2

Scransom · 技术社区 · 7 年前

我在r中运行了大量jags模型,使用 jags r2jags包的功能(它使用rjags包运行jags)。

控制台上有很多警告:

value out of range in 'lgamma'

打印这些警告似乎严重影响了计算时间。我该如何抑制?

警告打印为输出,而不是R警告。

我试过的不起作用的东西包括:

结束我的通话 try(..., silent = TRUE) , suppressWarnings , invisible ,或者 capture.output .
改变 jags.model 内部呼叫 贾格斯 到 jags.model(..., quiet = TRUE) .

这种现象也是 noted elsewhere ,我只想关闭它以减少从不必要的打印到控制台的大量计算负载。

有什么建议吗?

下面是一个很长但可重复的例子,基于 an example of the same issue on sourceforge . 很抱歉,这个长度,但我不能复制它在任何较小的玩具模型。我对这个特殊的模型不太在意,但它复制了这个问题 合理的 简单地说:

模型

cat('
model {
    K <- 1.1

    K.mvhypgeom <- exp( logfact(sum(n[])) - logfact(nMissing) - logfact( sum(n[]) - nMissing))

    p ~ dunif(0,1)

    for (t in 1:N) {
    X.missing[t] ~ dpois( missRate )
    }     

    nMissing ~ dsum(X.missing[1],X.missing[2],X.missing[3],X.missing[4],X.missing[5],X.missing[6],X.missing[7],X.missing[8],X.missing[9],X.missing[10])

    for (t in 1:N) {
    pX.missing[t] <- exp(logfact(n[t]) - logfact( X.missing[t]) - logfact( n[t] - X.missing[t]))
    ones2[t] ~ dbern(pX.missing[t]/K.mvhypgeom)
    }

    for (t in 1:N) {
    X[t] <- X.obs[t] + X.missing[t]

    likX[t] <- dbin( X[t], p, n[t])
    ones1[t] ~ dbern( likX[t] / K)    
    }
    }
  ',
    file = {example.model <- tempfile()},
    sep = ''
)

数据

simBinTS <- function(n, p , nMissing) {
  X.full <- X <- rbinom(N, size = n, prob = p)
  for (i in seq_len(nMissing)) {
    idx <- sample(1:N, size = 1, prob = X)
    X[idx] <- X[idx] - 1
  }
  return(data.frame(n = n, X = X, X.full = X.full))
}


N <- 10
p <- 0.3
set.seed(123)
n <- rpois(N, lambda = 30)
nMissing <- 10
missRate <- 1/10
ts <- simBinTS(p = p, n = n, nMissing = nMissing)
X.obs <- ts$X
n <- ts$n
X.full <- ts$X.full
ones1 <- rep(1,N)
ones2 <- rep(1,N)

jags.inits <- function(){
  list(X.missing = X.full-X.obs)
}

呼叫

library("R2jags")

jags(data = list("X.obs", "n", "N", "nMissing", "ones1", "ones2", "missRate"),
     inits = jags.inits,
     parameters.to.save = "p",
     model.file = example.model,
     n.chains = 3,
     n.iter = 1000,
     n.burnin = 500,
     n.thin = 1,
     progress.bar = "none")

产量 (大量重复的警告修剪-同样,这些是作为函数输出而不是作为警告消息打印的)

value out of range in 'lgamma'
value out of range in 'lgamma'
value out of range in 'lgamma'
value out of range in 'lgamma'
value out of range in 'lgamma'
value out of range in 'lgamma'
Inference for Bugs model at "D:\Users\fish\AppData\Local\Temp\RtmpWufTIC\file1614244456e1", fit using jags,
 3 chains, each with 1000 iterations (first 500 discarded)
 n.sims = 1500 iterations saved
         mu.vect sd.vect    2.5%     25%     50%     75%   97.5%  Rhat
p          0.331   0.027   0.280   0.312   0.330   0.348   0.388 1.006
deviance 812.379   2.761 808.165 810.345 811.941 814.103 818.729 1.007
         n.eff
p         1300
deviance   670

For each parameter, n.eff is a crude measure of effective sample size,
and Rhat is the potential scale reduction factor (at convergence, Rhat=1).

DIC info (using the rule, pD = var(deviance)/2)
pD = 3.8 and DIC = 816.2
DIC is an estimate of expected predictive error (lower deviance is better).

1 回复 | 直到 7 年前

1

3

Technophobe01 7 年前

问题在于 jags 把它打包 R2Jags 依靠。

JAGS使用 printf 不 fprintf 显示警告。jags不会把警告发送给 stderr ,它将警告发送到控制台而不是stderr。因此,R控制台无法筛选警告。

R2JAGS 依赖于贾格斯申请。我从sourceforge下载了jags源代码 JAGS-4.3.0 ,编译并安装库。这让我可以追踪代码并识别 jags 通过以下方式发出警告:

src/jrmath/lgamma.c:74 通过 ML_ERROR(ME_RANGE, "lgamma");

这就解决了

src/jrmath/nmath.h:138 通过 MATHLIB_WARNING(msg, s);

决定

src/jrmath/nmath.h:81
#define MATHLIB_WARNING(fmt,x) printf(fmt,x)

问题是 普林特 不使用 fprint(stderr,...) ,因此可以修补:

快速解决方案:

如果希望快速解决,可以下载源并应用以下修复程序:

$ diff nmath.h.orig nmath.h
81c81
< #define MATHLIB_WARNING(fmt,x)        printf(fmt,x)
---
> #define MATHLIB_WARNING(fmt,x)        fprintf(stderr,fmt,x)

现在您可以编译并安装jags库:

>./configure
>sudo make uninstall && sudo make install

完成后,我们可以卸载r2jags库,重新安装它,并使用带有stderr重定向的r cmd来抑制stderr…

R CMD ./50635735.R 2> /dev/null

代码示例

#!/usr/bin/env Rscript

library("R2jags")

source("./model.R") # Source Model
source("./simbits.R") # Source simBinTS code...

jags.data <- list("X.obs", "n", "N", "nMissing", "ones1", "ones2", "missRate")

model <- jags(data = jags.data,
              inits = jags.inits,
              parameters.to.save = "p",
              model.file = example.model,
              n.chains = 3,
              n.iter = 1000,
              n.burnin = 500,
              n.thin = 1,
              progress.bar = "none")

model

控制台输出:

未经改装的Jags

$ R CMD ./50635735.R 2> /dev/null
1 checking for pkg-config... /usr/local/bin/pkg-config
2 configure: Setting compile and link flags according to pkg-config
3 configure: Compile flags are -I/usr/local/include/JAGS
4 configure: Link flags are -L/usr/local/lib -ljags
5 checking for gcc... ccache clang
6 checking whether we are using the GNU C compiler... no
7 checking whether ccache clang accepts -g... no
8 checking for ccache clang option to accept ISO C89... unsupported
9 checking for jags_version in -ljags... yes
10 checking version of JAGS library... OK
11 configure: creating ./config.status
12 config.status: creating src/Makevars
13 configure: creating ./config.status
14 config.status: creating src/Makevars
15 config.status: creating R/unix/zzz.R
16 ccache clang++ -I"/usr/local/Cellar/r/3.5.0_1/lib/R/include" -DNDEBUG -I/usr/local/include/JAGS   -I/usr/local/opt/gettext/include -I/usr/
17 ccache clang++ -I"/usr/local/Cellar/r/3.5.0_1/lib/R/include" -DNDEBUG -I/usr/local/include/JAGS   -I/usr/local/opt/gettext/include -I/usr/
18 ccache clang++ -dynamiclib -Wl,-headerpad_max_install_names -undefined dynamic_lookup -single_module -multiply_defined suppress -L/usr/loc
19 Compiling model graph
20    Resolving undeclared variables
21    Allocating nodes
22 Graph information:
23    Observed stochastic nodes: 21
24    Unobserved stochastic nodes: 11
25    Total graph size: 174
26
27 Initializing model
28
29 value out of range in 'lgamma'
30 value out of range in 'lgamma'
31 value out of range in 'lgamma'
32 value out of range in 'lgamma'
...
...
...
10089 value out of range in 'lgamma'
10090 Inference for Bugs model at "/var/folders/md/03gdc4c14z18kbqwpfh4jdfc0000gp/T//Rtmp3P3FrI/file868156b0697", fit using jags,
10091  3 chains, each with 1000 iterations (first 500 discarded)
10092  n.sims = 1500 iterations saved
10093          mu.vect sd.vect    2.5%     25%     50%     75%   97.5%  Rhat n.eff
10094 p          0.333   0.027   0.281   0.315   0.332   0.350   0.391 1.003   590
10095 deviance 812.168   2.720 808.036 810.199 811.778 813.737 818.236 1.036    66
10096
10097 For each parameter, n.eff is a crude measure of effective sample size,
10098 and Rhat is the potential scale reduction factor (at convergence, Rhat=1).
10099
10100 DIC info (using the rule, pD = var(deviance)/2)
10101 pD = 3.6 and DIC = 815.8
10102
10103
10104
10105
10106
10107
10108
10109 BDIC is an estimate of expected predictive error (lower deviance is better).

改进的jags框架

$ R CMD ./50635735.R 2> /dev/null
checking for pkg-config... /usr/local/bin/pkg-config
configure: Setting compile and link flags according to pkg-config
configure: Compile flags are -I/usr/local/include/JAGS
configure: Link flags are -L/usr/local/lib -ljags
checking for gcc... ccache clang
checking whether we are using the GNU C compiler... no
checking whether ccache clang accepts -g... no
checking for ccache clang option to accept ISO C89... unsupported
checking for jags_version in -ljags... yes
checking version of JAGS library... OK
configure: creating ./config.status
config.status: creating src/Makevars
configure: creating ./config.status
config.status: creating src/Makevars
config.status: creating R/unix/zzz.R
ccache clang++ -I"/usr/local/Cellar/r/3.5.0_1/lib/R/include" -DNDEBUG -I/usr/local/include/JAGS   -I/usr/local/opt/gettext/include -I/usr/local/opt/readline/include -I/usr/local/include   -fPIC  -g -O2  -c jags.cc -o jags.o
ccache clang++ -I"/usr/local/Cellar/r/3.5.0_1/lib/R/include" -DNDEBUG -I/usr/local/include/JAGS   -I/usr/local/opt/gettext/include -I/usr/local/opt/readline/include -I/usr/local/include   -fPIC  -g -O2  -c parallel.cc -o parallel.o
ccache clang++ -dynamiclib -Wl,-headerpad_max_install_names -undefined dynamic_lookup -single_module -multiply_defined suppress -L/usr/local/opt/gettext/lib -L/usr/local/opt/readline/lib -L/usr/local/lib -L/usr/local/Cellar/r/3.5.0_1/lib/R/lib -L/usr/local/opt/gettext/lib -L/usr/local/opt/readline/lib -L/usr/local/lib -o rjags.so jags.o parallel.o -L/usr/local/lib -ljags -L/usr/local/opt/icu4c/lib -L/usr/local/lib -L/usr/local/Cellar/r/3.5.0_1/lib/R/lib -lR -lintl -Wl,-framework -Wl,CoreFoundation
Compiling model graph
   Resolving undeclared variables
   Allocating nodes
Graph information:
   Observed stochastic nodes: 21
   Unobserved stochastic nodes: 11
   Total graph size: 174

Initializing model

Inference for Bugs model at "/var/folders/md/03gdc4c14z18kbqwpfh4jdfc0000gp/T//RtmpI80TnH/file8e70516d6f34", fit using jags,
 3 chains, each with 1000 iterations (first 500 discarded)
 n.sims = 1500 iterations saved
         mu.vect sd.vect    2.5%     25%     50%     75%   97.5%  Rhat n.eff
p          0.333   0.027   0.281   0.315   0.332   0.350   0.391 1.003   590
deviance 812.168   2.720 808.036 810.199 811.778 813.737 818.236 1.036    66

For each parameter, n.eff is a crude measure of effective sample size,
and Rhat is the potential scale reduction factor (at convergence, Rhat=1).

DIC info (using the rule, pD = var(deviance)/2)
pD = 3.6 and DIC = 815.8
DIC is an estimate of expected predictive error (lower deviance is better).

长期解决方案

提交一个bug并通过sourceforge建议修复。