Skip to content

What should dnorm(0, 0, -Inf) return?

5 messages · Stephen D. Weigand, Jiefei Wang, Martin Maechler +1 more

Original
Stephen D. Weigand
# 
Hi,
Apropos of a recent Inf question, I've previously wondered if dnorm "does the right thing" with

  dnorm(0, 0, -Inf)

which gives zero. Should that be zero or NaN (or NA)? 

The help says "'sd < 0' is an error and returns 'NaN'" and since -Inf < 0 is TRUE, then... is this a bug?

Thank you,
Stephen
Rochester, MN USA
Jiefei Wang
# 
Good question, I cannot speak for R's developers but I would like to
provide some information on the problem. Here are the first few lines of
the dnorm function located at src\nmath\dnorm.c:

```
double dnorm4(double x, double mu, double sigma, int give_log)
{
#ifdef IEEE_754
    if (ISNAN(x) || ISNAN(mu) || ISNAN(sigma))
return x + mu + sigma;
#endif
    if(!R_FINITE(sigma)) return R_D__0;
    if(!R_FINITE(x) && mu == x) return ML_NAN;/* x-mu is NaN */
    if (sigma <= 0) {
    if (sigma < 0) ML_ERR_return_NAN;
        /* sigma == 0 */
        return (x == mu) ? ML_POSINF : R_D__0;
    }
    ....
}
```

You can clearly see where the problem is. I think either the document or
the code needs a modification.

Best,
Jiefei

On Sat, Dec 7, 2019 at 5:05 PM Weigand, Stephen D. via R-devel <
r-devel at r-project.org> wrote:

            

      
      
      
    

        

      
      
    

        

  
  


  


      

    

    

      
      

        


  

        

      Peter Dalgaard
    

    

      
      #
    

  


  

      
Yes, that looks like a bug and an easily fixable one too.

However, I spy another issue: Why do we check the !R_FINITE(x) && mu == x before checking for sd < 0 ? The difference is whether we 

return ML_NAN;
or
ML_ERR_return_NAN;

but surely negative sd should always be an error?

I'd be inclined to do

    if (sigma < 0) ML_ERR_return_NAN;
    if(!R_FINITE(sigma)) return R_D__0;
    if(!R_FINITE(x) && mu == x) return ML_NAN;/* x-mu is NaN */
    if (sigma == 0) 
        return (x == mu) ? ML_POSINF : R_D__0;
    x = (x - mu) / sigma;


(Ping Martin...)

-pd

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Martin Maechler
    

    

      
      #
    

  


  

          

      
      
      
    

        
> Yes, that looks like a bug and an easily fixable one too.

agreed.

    > However, I spy another issue: Why do we check the
    > !R_FINITE(x) && mu == x before checking for sd < 0 ? The
    > difference is whether we

    > return ML_NAN; or ML_ERR_return_NAN;

    > but surely negative sd should always be an error?

    > I'd be inclined to do

    > if (sigma < 0) ML_ERR_return_NAN;
    > if(!R_FINITE(sigma)) return R_D__0;
    > if(!R_FINITE(x) && mu == x) return ML_NAN;/* x-mu is NaN */
    > if (sigma == 0) 
    >    return (x == mu) ? ML_POSINF : R_D__0;
    > x = (x - mu) / sigma;


    > (Ping Martin...)

I think you are spot on, Peter.
All of this code has a longish history, with incremental border
case improvements.
Let's hope (somewhat unrealistically) this is the last one for
dnorm().

NB: dlnorm() and some of the gamma/chisq/.. may need a
    similar adjustment

Lastly, regression tests for this
(either in  tests/d-p-q-r-tests.{R,Rout.save}
 or easier in reg-tests-1d.R)  should be added too.

    > -pd

        
>> On 7 Dec 2019, at 23:40 , Wang Jiefei <szwjf08 at gmail.com> wrote:

        
>> 
    >> Good question, I cannot speak for R's developers but I would like to
    >> provide some information on the problem. Here are the first few lines of
    >> the dnorm function located at src\nmath\dnorm.c:
    >> 
    >> ```
    >> double dnorm4(double x, double mu, double sigma, int give_log)
    >> {
    >> #ifdef IEEE_754
    >> if (ISNAN(x) || ISNAN(mu) || ISNAN(sigma))
    >> return x + mu + sigma;
    >> #endif
    >> if(!R_FINITE(sigma)) return R_D__0;
    >> if(!R_FINITE(x) && mu == x) return ML_NAN;/* x-mu is NaN */
    >> if (sigma <= 0) {
    >> if (sigma < 0) ML_ERR_return_NAN;
    >> /* sigma == 0 */
    >> return (x == mu) ? ML_POSINF : R_D__0;
    >> }
    >> ....
    >> }
    >> ```
    >> 
    >> You can clearly see where the problem is. I think either the document or
    >> the code needs a modification.
    >> 
    >> Best,
    >> Jiefei
    >> 
    >> On Sat, Dec 7, 2019 at 5:05 PM Weigand, Stephen D. via R-devel <

        
>> r-devel at r-project.org> wrote:

        
>> 
    >>> Hi,
    >>> Apropos of a recent Inf question, I've previously wondered if dnorm "does
    >>> the right thing" with
    >>> 
    >>> dnorm(0, 0, -Inf)
    >>> 
    >>> which gives zero. Should that be zero or NaN (or NA)?
    >>> 
    >>> The help says "'sd < 0' is an error and returns 'NaN'" and since -Inf < 0
    >>> is TRUE, then... is this a bug?
    >>> 
    >>> Thank you,
    >>> Stephen
    >>> Rochester, MN USA
    >>> 
    >>> ______________________________________________
    >>> R-devel at r-project.org mailing list
    >>> https://stat.ethz.ch/mailman/listinfo/r-devel
    >>> 
    >> 
    >> [[alternative HTML version deleted]]
    >> 
    >> ______________________________________________
    >> R-devel at r-project.org mailing list
    >> https://stat.ethz.ch/mailman/listinfo/r-devel

    > -- 
    > Peter Dalgaard, Professor,
    > Center for Statistics, Copenhagen Business School
    > Solbjerg Plads 3, 2000 Frederiksberg, Denmark
    > Phone: (+45)38153501
    > Office: A 4.23
    > Email: pd.mes at cbs.dk  Priv: PDalgd at gmail.com

  
  


  


      

    

    

      
      

        


  

        

      Peter Dalgaard
    

    

      
      #
    

  


  

      
I have committed a fix for r-devel (dnorm only).

-pd