[R] number of decimal

[David Winsemius]

On Jan 28, 2010, at 12:08 PM, Marc Schwartz wrote:

>
> On Jan 28, 2010, at 10:04 AM, David Winsemius wrote:
>
>>
>> On Jan 28, 2010, at 10:55 AM, Marc Schwartz wrote:
>>
>>> Ivan,
>>>
>>> The default behavior for print()ing objects to the console in an R  
>>> session is via the use of the print.* methods. For real numerics,  
>>> print.default() is used and the format is based upon the number of  
>>> significant digits, not the number of decimal places. There is  
>>> also an interaction with par("scipen"), which influences when  
>>> scientific notation is used. See ?print.default for more  
>>> information on defaults and behavior, taking note of the 'digits'  
>>> argument, which is influenced by options("digits").
>>>
>>> Importantly, you need to differentiate between how R stores  
>>> numeric real values and how it displays or prints them.  
>>> Internally, R stores real numbers using a double precision data  
>>> type by default.
>>>
>>> The internal storage is not truncated by default and is stored to  
>>> full precision for doubles, within binary representation limits.  
>>> You can of course modify the values using functions such as  
>>> round() or truncate(), etc. See ?round for more information.
>>>
>>> For display, Peter has already pointed you to sprintf() and  
>>> related functions, which allow you to format output for "pretty  
>>> printing" to things like column aligned tables and such. Those do  
>>> not however, affect the default output to the R console.
>>
>> If one alters print.default, one can get different behavior, for  
>> instance:
>>
>> print.default <- function (x, digits = NULL, quote = TRUE, na.print  
>> = NULL, print.gap = NULL,
>>   right = FALSE, max = NULL, useSource = TRUE, ...)
>> {if (is.numeric(x)) {x <- as.numeric(sprintf("%7.3f", x))}
>>   noOpt <- missing(digits) && missing(quote) && missing(na.print) &&
>>       missing(print.gap) && missing(right) && missing(max) &&
>>       missing(useSource) && length(list(...)) == 0L
>>   .Internal(print.default(x, digits, quote, na.print, print.gap,
>>       right, max, useSource, noOpt))
>> }
>>
>> This will have the requested effect for numeric vectors, but does  
>> not seem to be altering the behavior of print.data.frame().
>>
>>> print(ac2)
>>      score pt times trt
>> 1  28.825139  1     0   1
>> 2  97.458521  1     3   1
>> 3  26.217289  1     6   1
>> 4  80.636507  2     0   1
>> 5  99.729364  2     3   1
>> 6  85.812312  2     6   1
>> 7   2.515870  3     0   1
>> 8   3.893545  3     3   1
>> 9  55.666848  3     6   1
>> 10 21.966027  4     0   1
>>> print(ac2$score)
>> [1] 28.825 97.459 26.217 80.637 99.729 85.812  2.516  3.894 55.667  
>> 21.966
>>
>
>
> David,
>
> The issue there is that when printing the vector, you are using  
> print.default() directly, so you get the desired result with a  
> numeric vector.

Thanks, Marc;

I do understand. I had been hoping that there might be a "final common  
pathway" to use a biochemistry analogy, at least for numeric objects,  
but it appears not.

-- 
David.

>
> When you print the data frame, internally print.data.frame() calls  
> format.data.frame(), which then internally uses format() on a column- 
> by-column basis and there is the rub. format() brings you back to  
> using significant digits on numeric vectors and of course returns a  
> character vector. By the time the output is actually print()ed to  
> the console, the original data frame has been converted to a  
> formatted character matrix and that is what gets printed.
>
>> str(format.data.frame(ac2))
> 'data.frame':	10 obs. of  4 variables:
> $ score:Class 'AsIs'  chr [1:10] "28.825139" "97.458521" "26.217289"  
> "80.636507" ...
> $ pt   :Class 'AsIs'  chr [1:10] "1" "1" "1" "2" ...
> $ times:Class 'AsIs'  chr [1:10] "0" "3" "6" "0" ...
> $ trt  :Class 'AsIs'  chr [1:10] "1" "1" "1" "1" ...
>
>
>> str(format.data.frame(ac2, digits = 2))
> 'data.frame':	10 obs. of  4 variables:
> $ score:Class 'AsIs'  chr [1:10] " 28.8" " 97.5" " 26.2" " 80.6" ...
> $ pt   :Class 'AsIs'  chr [1:10] "1" "1" "1" "2" ...
> $ times:Class 'AsIs'  chr [1:10] "0" "3" "6" "0" ...
> $ trt  :Class 'AsIs'  chr [1:10] "1" "1" "1" "1" ...
>
>
> This is why changing print.default() by itself is not sufficient.  
> Other object classes are formatted and printed in varying ways and  
> print methods have been defined for them which may not use it  
> directly.
>
> HTH,
>
> Marc Schwartz
>

David Winsemius, MD
Heritage Laboratories
West Hartford, CT