Before someone closes the question, yes, there are many questions that seem similar, but so far I haven't found one with this exact weird problem that seems to go away only sometimes.
I had an odd Fortran error while trying to make a module for linear regression.
The module is named "LSQregression" and the main program "LSQAdvertising". Compiling it using the following gfortran command works:
gfortran ../LSQregression.f90 LinearAdvertising.f90 -llapack -o LinAd
However, I'd like to be able to turn my module into a .o file and link it with whatever program I may need instead of compiling again every time. So I tried to do that:
gfortran -c ../LSQregression.f90
And then link it with my program file like that:
gfortran ../LSQregression.o LinearAdvertising.f90 -llapack -o LinAd
I also tried also turning the program into a .o file. Neither works, they both return the following error:
/usr/bin/ld: /tmp/ccf7T1aj.o: in function `MAIN__':
LinearAdvertising.f90:(.text+0x1f8b): undefined reference to `__lsqregression_MOD_lsqestimate_simple'
collect2: error: ld returned 1 exit status
It refers to the following function in the module that is being called here inside the program:
print *, LSQestimate(test,LSQbeta(inputs,sales,(/1,2/)) )
The function itself is defined as part of an interface:
interface LSQestimate
procedure LSQestimate_simple, LSQestimate_using
end interface LSQestimate
real(8) function LSQestimate_simple(X, beta)
implicit none
real(8), dimension(:,:) :: X, beta
real(8) :: boundary, Y
integer :: i, Xlen, betalen
Xlen = size(X, 1)
betalen = size(beta, 1)
if (Xlen + 1 .ne. betalen) stop "incompatible beta and X"
Y = beta(1,1)
do i = 1, Xlen
Y = Y + beta(i+1, 1)*X(i,1)
end do
LSQestimate_simple = Y
end function LSQestimate_simple
It's very odd because I've done the same thing with another function and it seems to work fine, and the problem only happens when I turn the module into a .o file first, and goes away if I try to directly compile the .f90 file... I can't figure out why one works and the other doesn't.
EDIT: Someone told me to do nm ../LSQregression.o | grep -i regression . I have no idea what that does but I did it so here are the results if it helps at all:
0000000000002768 T __lsqregression_MOD_inv
0000000000000b85 T __lsqregression_MOD_lsqbeta_simple
000000000000035c T __lsqregression_MOD_lsqbeta_using
0000000000000000 T __lsqregression_MOD_lsqdecision
Oddly enough the name of the function in question doesn't even appear here.
EDIT 2: Decided to post the entire code of the module:
module LSQregression
implicit none
public :: LSQbeta, LSQestimate
interface LSQbeta
procedure LSQbeta_simple, LSQbeta_using
end interface LSQbeta
interface LSQestimate
procedure LSQestimate_simple, LSQestimate_using
end interface LSQestimate
contains
function inv(A) result(Ainv)
implicit none
real(8), dimension(:,:), intent(in) :: A
real(8), dimension(size(A,1),size(A,2)) :: Ainv
real(8), dimension(size(A,1)) :: work ! work array for LAPACK
integer, dimension(size(A,1)) :: ipiv ! pivot indices
integer :: n, info
! External procedures defined in LAPACK
external DGETRF
external DGETRI
! Store A in Ainv to prevent it from being overwritten by LAPACK
Ainv = A
n = size(A,1)
! DGETRF computes an LU factorization of a general M-by-N matrix A
! using partial pivoting with row interchanges.
call DGETRF(n, n, Ainv, n, ipiv, info)
if (info /= 0) then
stop 'Matrix is numerically singular!'
end if
! DGETRI computes the inverse of a matrix using the LU factorization
! computed by DGETRF.
call DGETRI(n, Ainv, n, ipiv, work, n, info)
if (info /= 0) then
stop 'Matrix inversion failed!'
end if
end function inv
!----------------------------------------------------------------------
function LSQbeta_simple(Xold, Y) result(beta)
real(8), dimension(:,:) :: Xold, Y
real(8), dimension(size(Xold,1), size(Xold,2)+1) :: X
real(8), dimension(:,:), allocatable :: beta, XTX, IXTX, pinv
integer :: rowsX, colsX, rowsY, colsY,i
rowsX = size(X(:,1))
colsX = size(X(1,:))
rowsY = size(Y(:,1))
colsY = size(Y(1,:))
if (colsY .ne. 1) stop 'Inappropriate y'
if (rowsY .ne. rowsX) stop "Y and X rows don't match"
!---X-ify-----------
X(:,1) = 1.0 !includes 1 in the first column
do i = 2, colsX
X(:,i) = Xold(:,i-1)
end do
!--------------------
allocate(XTX(colsX,colsX))
allocate(IXTX(colsX,colsX))
allocate(pinv(colsX,rowsY))
allocate(beta(colsX,1))
XTX = matmul(transpose(X),X)
IXTX = inv(XTX)
pinv = matmul(IXTX, transpose(X))
beta = matmul(pinv, Y)
deallocate(pinv)
deallocate(XTX)
deallocate(IXTX)
return
deallocate(beta)
end function LSQbeta_simple
!--------------------------------------------
function LSQbeta_using(Xold, Y, indexes) result(beta)
implicit none
real(8), dimension(:,:) :: Xold, Y
integer, dimension(:) :: indexes
real(8), dimension(size(indexes)+1,1) :: beta
real(8), dimension(size(Xold,1),size(indexes)) :: X
integer :: indLen,i
indLen = size(indexes)
do i = 1, indLen
X(:,i) = Xold(:,indexes(i))
end do
beta = LSQbeta_simple(X,Y)
end function LSQbeta_using
real(8) function LSQestimate_simple(X, beta)
implicit none
real(8), dimension(:,:) :: X, beta
real(8) :: boundary, Y
integer :: i, Xlen, betalen
Xlen = size(X, 1)
betalen = size(beta, 1)
if (Xlen + 1 .ne. betalen) stop "incompatible beta and X"
Y = beta(1,1)
do i = 1, Xlen
Y = Y + beta(i+1, 1)*X(i,1)
end do
LSQestimate_simple = Y
end function LSQestimate_simple
real(8) function LSQestimate_using()
LSQestimate_using = 1.0D0
end function LSQestimate_using
logical function LSQdecision(X, beta, boundary)
implicit none
real(8), dimension(:,:) :: X, beta
real(8) :: boundary
LSQdecision = LSQestimate(X, beta) > boundary
end function LSQdecision
end module LSQregression
