Continuing on with a loop when there's a failure

On Jul 13, 2010, at 9:24 AM, Josh B wrote:

In my opinion the try and tryCatch commands are written and  
documented rather poorly. Thus I am not sure what to program exactly.

Didn't  you see the "silent" parameter? Its seems to be documented  
fairly clearly to me.

The testing of try at the console is not going to be very  
illuminating, since it really only has value within a function that  
you want to continue despite an error. try() WILL provide that  
facility but _you_ need to decide what you do with the information  
it returns, which in the case of its use with the default  
silent=FALSE is just the error message itself.

For instance, I could query mod.poly3 and use an if/then statement  
to proceed,

So why didn't you? A good result would be signaled by:

-- 
David.

but querying mod.poly3 is weird. For instance, here's the output  
when it fails:

mod.poly3 <- try(lrm(x[,2] ~ pol(x1, 3) + pol(x2, 3), data=x))

Error in fitter(X, Y, penalty.matrix = penalty.matrix, tol = tol,  
weights = weights,  :
 NA/NaN/Inf in foreign function call (arg 1)

mod.poly3

[1] "Error in fitter(X, Y, penalty.matrix = penalty.matrix, tol =  
tol, weights = weights,  : \n  NA/NaN/Inf in foreign function call  
(arg 1)\n"
attr(,"class")
[1] "try-error"

...and here's the output when it succeeds:

mod.poly3 <- try(lrm(x[,1] ~ pol(x1, 3) + pol(x2, 3), data=x))
mod.poly3

Logistic Regression Model

lrm(formula = x[, 1] ~ pol(x1, 3) + pol(x2, 3), data = x)


Frequencies of Responses
bagels donuts
   10      5

      Obs  Max Deriv Model L.R.       d.f.          P          C
       15      4e-04       3.37          6     0.7616       0.76
      Dxy      Gamma      Tau-a         R2      Brier          g
     0.52       0.52      0.248      0.279      0.183      1.411
       gr         gp
      4.1      0.261

         Coef     S.E.    Wald Z P
Intercept -5.68583 5.23295 -1.09  0.2772
x1         1.87020 2.14635  0.87  0.3836
x1^2      -0.42494 0.48286 -0.88  0.3788
x1^3       0.02845 0.03120  0.91  0.3618
x2         3.49560 3.54796  0.99  0.3245
x2^2      -0.94888 0.82067 -1.16  0.2476
x2^3       0.06362 0.05098  1.25  0.2121

...so what exactly would I query to design my if/then statement?

From: David Winsemius <dwinsemius at comcast.net>
To: David Winsemius <dwinsemius at comcast.net>
Cc: Josh B <joshb41 at yahoo.com>; R Help <r-help at r-project.org>
Sent: Tue, July 13, 2010 9:09:04 AM
Subject: Re: [R] Continuing on with a loop when there's a failure


On Jul 13, 2010, at 9:04 AM, David Winsemius wrote:

On Jul 13, 2010, at 8:47 AM, Josh B wrote:

Thanks again, David.

...but, alas, I still can't get it work!

(BTW, it did work.)

Here's what I'm trying now:

for (i in 1:2) {
   mod.poly3 <- try(lrm(x[,i] ~ pol(x1, 3) + pol(x2, 3), data=x))
   results[1,i] <- anova(mod.poly3)[1,3]
}

You need to do some programming.

(Or I suppose you could wrap both the lrm and the anova calls in  
try.)

You did not get an error from the lrm but rather from the anova  

call because you tried to give the results of the try function to  
anova without first checking to see if an error had occurred.

--David.

Here's what happens (from the console):

Error in fitter(X, Y, penalty.matrix = penalty.matrix, tol =  

tol, weights = weights,  :

 NA/NaN/Inf in foreign function call (arg 1)
Error in UseMethod("anova") :
 no applicable method for 'anova' applied to an object of class  

"try-error"

...so I still can't make my results matrix. Could I ask you for  

some specific code to make this work? I'm not that familiar with  
the syntax for try or tryCatch, and the help files for them are  
pretty bad, in my humble opinion.

I should clarify that I actually don't care about the failed  

runs per se. I just want R to keep going in spite of them and give  
me my results matrix.

From: David Winsemius <dwinsemius at comcast.net>
To: Josh B <joshb41 at yahoo.com>
Cc: R Help <r-help at r-project.org>
Sent: Mon, July 12, 2010 8:09:03 PM
Subject: Re: [R] Continuing on with a loop when there's a failure


On Jul 12, 2010, at 6:18 PM, Josh B wrote:

Hi R sages,

Here is my latest problem. Consider the following toy example:

x <- read.table(textConnection("y1 y2 y3 x1 x2
indv.1 bagels donuts bagels 4 6
indv.2 donuts donuts donuts 5 1
indv.3 donuts donuts donuts 1 10
indv.4 donuts donuts donuts 10 9
indv.5 bagels donuts bagels 0 2
indv.6 bagels donuts bagels 2 9
indv.7 bagels donuts bagels 8 5
indv.8 bagels donuts bagels 4 1
indv.9 donuts donuts donuts 3 3
indv.10 bagels donuts bagels 5 9
indv.11 bagels donuts bagels 9 10
indv.12 bagels donuts bagels 3 1
indv.13 donuts donuts donuts 7 10
indv.14 bagels donuts bagels 2 10
indv.15 bagels donuts bagels 9 6"), header = TRUE)

I want to fit a logistic regression of y1 on x1 and x2. Then I  

want to run a

logistic regression of y2 on x1 and x2. Then I want to run a  

logistic regression

of y3 on x1 and x2. In reality I have many more Y columns than  

simply "y1,"

"y2," and "y3," so I must design a loop. Notice that y2 is  

invariant and thus it

will fail. In reality, some y columns will fail for much more  

subtle reasons.

Simply screening my data to eliminate invariant columns will  

not eliminate the

problem.

What I want to do is output a piece of the results from each  

run of the loop to

a matrix. I want the to try each of my y columns, and not give  

up and stop

running simply because a particular y column is bad. I want it  

to give me "NA"

or something similar in my results matrix for the bad y  

columns, but I want it

to keep going give me good data for the good y columns.

For instance:
results <- matrix(nrow = 1, ncol = 3)
colnames(results) <- c("y1", "y2", "y3")

for (i in 1:2) {
mod.poly3 <- lrm(x[,i] ~ pol(x1, 3) + pol(x2, 3), data=x)
results[1,i] <- anova(mod.poly3)[1,3]
}

If I run this code, it gives up when fitting y2 because the y2  

is bad. It

doesn't even try to fit y3. Here's what my console shows:

results

           y1 y2 y3
[1,] 0.6976063 NA NA

As you can see, it gave up before fitting y3, which would have  

worked.

How do I force my code to keep going through the loop, despite  

the rotten apples

it encounters along the way?

?try

http://cran.r-project.org/doc/FAQ/R-FAQ.html#How-can-I-capture-or-ignore-errors-in-a-long-simulation_003f

(Doesn't only apply to simulations.)

Exact code that gets the job done is what I am
interested in. I am a post-doc -- I am not taking any classes.  

I promise this is

not a homework assignment!

--
David Winsemius, MD
West Hartford, CT

David Winsemius, MD
West Hartford, CT

David Winsemius, MD
West Hartford, CT

David Winsemius, MD
West Hartford, CT