[Rd] How does .Fortran "dqrls" work?

[cberry at tajo.ucsd.edu]

yangleicq <yanglei_cq at 126.com> writes:

> Hi, all.
> I want to write some functions like glm() so i studied it.
> In glm.fit(), it calls a fortran subroutine named  "dqrfit" to compute least
> squares solutions
>  to the system
>               x * b = y
>
> To learn how "dqrfit" works, I just follow how glm() calls "dqrfit" by my
> own example, my codes are given below:
>
>>  qr <-
>> matrix(c(4.17,5.58,5.18,6.11,4.50,4.61,5.17,4.53,5.33,5.14,2.3,1.7,1.3,1.7,1.7,1.6,1,1.7,1.7,1.7),ncol=2)
>> qr 
>       [,1] [,2]
>  [1,] 4.17  2.3
>  [2,] 5.58  1.7
>  [3,] 5.18  1.3
>  [4,] 6.11  1.7
>  [5,] 4.50  1.7
>  [6,] 4.61  1.6
>  [7,] 5.17  1.0
>  [8,] 4.53  1.7
>  [9,] 5.33  1.7
> [10,] 5.14  1.7
>> n=10
>>  p=2
>>  y <- c(4.81,4.17,4.41,3.59,5.87,3.83,6.03,4.89,4.32,4.69)
>>  ny=1L
>>  tol=1e-07
>>  coefficients=double(p)
>>  residuals=double(n)
>>  effects=double(n)
>>  rank=integer(1L)
>>  pivot=1:n
>>  qraux=double(n)
>>  work=double(2*n)
>>  
>>  
>>  
>>  fittt<-.Fortran("dqrls", qr =qr, n = n, 
> +                 p = p, y = y, ny = ny, tol = tol,
> coefficients=coefficients,
> +                 residuals = residuals, effects = effects, 
> +                 rank = rank, pivot = pivot, qraux = qraux, 
> +                 work = work, PACKAGE = "base")
>>  
>>  fittt$coefficients
> [1] 0 0

You have the args for .Fortran wrong. Try:

> fargs <- structure(list("dqrls", qr = structure(c(1, 1, 1, 1, 1, 1, 1, 
+ 1, 1, 1, 4.17, 5.58, 5.18, 6.11, 4.5, 4.61, 5.17, 4.53, 5.33, 
+ 5.14, 2.3, 1.7, 1.3, 1.7, 1.7, 1.6, 1, 1.7, 1.7, 1.7), .Dim = c(10L, 
+ 3L)), n = 10L, p = 3L, y = c(4.81, 4.17, 4.41, 3.59, 5.87, 3.83, 
+ 6.03, 4.89, 4.32, 4.69), ny = 1L, tol = 1e-11, coefficients = c(0, 
+ 0, 0), residuals = c(0, 0, 0, 0, 0, 0, 0, 0, 0, 0), effects = c(0, 
+ 0, 0, 0, 0, 0, 0, 0, 0, 0), rank = 0L, pivot = 1:3, qraux = c(0, 
+ 0, 0), work = c(0, 0, 0, 0, 0, 0), PACKAGE = "base"), .Names = c("", 
+ "qr", "n", "p", "y", "ny", "tol", "coefficients", "residuals", 
+ "effects", "rank", "pivot", "qraux", "work", "PACKAGE"))
> do.call(.Fortran,fargs)$coef
[1] 11.176571 -0.883272 -1.262772
>

TIP: It often helps to use something like

     debug(function.calling.Fortran) 

and then step thru the function till the call you want to study is
invoked. Then inspect the inputs one-by-one and tinker with them and
recall the function or save them via 

         dput( list(...) , file="fargs" ) 

so you can later invoke the function as above.

HTH,

Chuck 


>
> but when i use lm() which also calls "dqrls" internally to fit this model,
> it gives reasonable result.
>
>> lm(y~qr)
>
> Call:
> lm(formula = y ~ qr)
>
> Coefficients:
> (Intercept)          qr1          qr2  
>     11.1766      -0.8833      -1.2628  
>
>
> when I change the coefficients to be c(1,1), the output from "dqrls", 
> fittt$coefficients also equals to c(1,1). That means the  .Fortran("dqrls",
> qr=qr,n=n,p=p,...) did nothing to the coefficients! I don't know why, is
> there anything I did wrong or missed?  How can I get the result from "dqrls"
> as what lm() or glm() gets from "dqrls"?
>
> Thanks in advance. Best Regards.
>
>
>
> --
> View this message in context: http://r.789695.n4.nabble.com/How-does-Fortran-dqrls-work-tp4588973p4588973.html
> Sent from the R devel mailing list archive at Nabble.com.
>

-- 
Charles C. Berry                            Dept of Family/Preventive Medicine
cberry at ucsd edu			    UC San Diego
http://famprevmed.ucsd.edu/faculty/cberry/  La Jolla, San Diego 92093-0901