qml is a library for statistics, linear algebra, and optimization in kdb+. It provides an interface between the q programming language and numerical libraries such as LAPACK.
The current version of the library is 0.5.2 (track updates). This is a development branch. For precompiled binaries see version 0.2.1.
The latest qml 0.5 wraps up some work I started over three years ago, right after releasing qml 0.2.1. The idea is for the optimization routines .qml.min and .qml.conmin to become a single uniform interface to different underlying libraries such as CONMAX and NLopt.
NLopt provides qml with derivativefree algorithms for unconstrained and constrained optimization. While CONMAX, the only algorithm available previously, also works without explicitly specified gradient functions, it has to approximate them with central differences.
Welcome qml 0.4. Development continues on GitHub.
I am working on a new version of QML.
The interface will be the same as in version 0.3, including the recently added mm, ms, mls and mlsq functions, while the configuration will be more like version 0.2, with BLAS used by default and precompiled binaries provided.
It will be compatible with kdb+ 3.0, testing will be more thorough than ever, and there will be an option to link against a different implementation of BLAS, such as ATLAS.
In the mean time, check out this patch from Kim Tang for kdb+ 3.0 support.
Also, if you want to build QML 0.3.10, the particular development snapshot of ATLAS that it uses is no longer available for download, so I'm hosting it here.
atlas3.9.46.tar.bz2 (5.31 MB) 
Just verified that it still builds on 32bit Debian.
QML 0.2.1 is still the stable version. It does not have the matrix performance of ATLAS, but it is easier to compile and includes binaries for 32bit platforms.
Since it was released some of the files on the Netlib site changed, and QML 0.2.1 no longer compiles with them. I'm hosting the original files here. Place them in the download subdirectory before running make.
clapack.tgz (6.89 MB) f2c.tar.gz (240.00 KB) 
qml is free software with a BSDstyle license. It is provided in the hope that it will be useful, but with absolutely no warranty. See the file LICENSE.txt for details.
qml is linked against several other libraries. The copyrights and licenses for these libraries are also listed in LICENSE.txt.

To compile and install from source code, run
./configure make make test make install
To install a precompiled binary, copy qml.q into the same directory as q.k, and copy qml.dll or qml.so into the same directory as q.exe or q. Then run test.q.
Load with
q)\l qml.q
All functions are in the .qml namespace. Numerical arguments are automatically converted into floatingpoint. Matrixes are in the usual rowmajor layout (lists of row vectors). Complex numbers are represented as pairs of their real and imaginary parts.
q).qml.nicdf .25 .5 .975 / normal distribution quantiles 0.6744898 0 1.959964 q).qml.mchol (1 2 1;2 5 4;1 4 6) / Cholesky decomposition 1 2 1 0 1 2 0 0 1 q).qml.poly 2 9 16 15 / solve 2x^39x^2+16x15=0 2.5 1 1.414214 1 1.414214 q).qml.mlsq[(1 1;1 2;1 3;1 4);11 2 3 4] / fit line 14 5f q).qml.conmin[{x*y+1};{1(x*x)+y*y};0 0] / minimize x(y+1) s.t. x^2+y^2<=1 0.8660254 0.5
There are more examples here and in my directory at code.kx.com.
pi  pi 
e  e 
eps  smallest representable step from 1. 
sin[x]  sine 
cos[x]  cosine 
tan[x]  tangent 
asin[x]  arcsine 
acos[x]  arccosine 
atan[x]  arctangent 
atan2[x;y]  atan[x%y] 
sinh[x]  hyperbolic sine 
cosh[x]  hyperbolic cosine 
tanh[x]  hyperbolic tangent 
asinh[x]  hyperbolic arcsine 
acosh[x]  hyperbolic arccosine 
atanh[x]  hyperbolic arctangent 
exp[x]  exponential 
expm1[x]  exp[x]1 
log[x]  logarithm 
log10[x]  base10 logarithm 
logb[x]  extract binary exponent 
log1p[x]  log[1+x] 
pow[a;x]  exponentiation 
sqrt[x]  square root 
cbrt[x]  cube root 
hypot[x;y]  sqrt[pow[x;2]+pow[y;2]] 
floor[x]  round downward 
ceil[x]  round upward 
fabs[x]  absolute value 
fmod[x;y]  remainder of x%y 
erf[x]  error function 
erfc[x]  complementary error function 
lgamma[x]  log of absolute value of gamma function 
gamma[x]  gamma function 
beta[x;y]  beta function 
pgamma[a;x]  lower incomplete gamma function (a>0) 
pgammac[a;x]  upper incomplete gamma function (a>0) 
pgammar[a;x]  regularized lower incomplete gamma function (a>0) 
pgammarc[a;x]  regularized upper incomplete gamma function (a>0) 
ipgammarc[a;p]  inverse complementary regularized incomplete gamma function (a>0, p≥0.5) 
pbeta[a;b;x]  incomplete beta function (a,b>0) 
pbetar[a;b;x]  regularized incomplete beta function (a,b>0) 
ipbetar[a;b;p]  inverse regularized incomplete beta function (a,b>0) 
j0[x]  order 0 Bessel function 
j1[x]  order 1 Bessel function 
y0[x]  order 0 Bessel function of the second kind 
y1[x]  order 1 Bessel function of the second kind 
ncdf[x]  CDF of normal distribution 
nicdf[p]  its inverse 
c2cdf[k;x]  CDF of chisquared distribution (k≥1) 
c2icdf[k;p]  its inverse 
stcdf[k;x]  CDF of Student's tdistribution (natural k) 
sticdf[k;p]  its inverse 
fcdf[d1;d2;x]  CDF of Fdistribution (d1,d2≥1, x≥0) 
ficdf[d1;d2;p]  its inverse 
gcdf[k;th;x]  CDF of gamma distribution 
gicdf[k;th;x]  its inverse 
bncdf[k;n;p]  CDF of binomial distribution 
bnicdf[k;n;x]  its inverse for p parameter (k<n) 
pscdf[k;lambda]  CDF of Poisson distribution 
psicdf[k;p]  its inverse for lambda 
smcdf[n;e]  CDF for onesided KolmogorovSmirnov test 
smicdf[n;e;x]  its inverse 
kcdf[x]  CDF for Kolmogorov distribution 
kicdf[p]  its inverse (p≥1e8) 
diag[diag]  make diagonal matrix 
mdim[matrix]  number of (rows; columns) 
mdiag[matrix]  extract main diagonal 
mdet[matrix]  determinant 
mrank[matrix]  rank 
minv[matrix]  inverse 
mpinv[matrix]  pseudoinverse 
mm[A;B]  multiply 
ms[A;B]  solve B=A mm X, A is triangular 
mev[matrix]  (eigenvalues; eigenvectors) sorted by decreasing modulus 
mchol[matrix]  Cholesky factorization upper matrix 
mqr[matrix]  QR factorization: (Q; R) 
mqrp[matrix]  QR factorization with column pivoting: (Q; R; P), matrix@\:P=Q mm R 
mlup[matrix]  LUP factorization with row pivoting: (L; U; P), matrix[P]=L mm U 
msvd[matrix]  singular value decomposition: (U; Sigma; V) 
mkron[A;B]  Kronecker product 
poly[coef]  roots of a polynomial (highestdegree coefficient first, can be complex) 
mls[A;B]  solve B=A mm X  
mlsx[opt;A;B]  mls[] with options
 
mlsq[A;B]  solve min BA mm X  
mlsqx[opt;A;B]  mlsq[] with options

root[f;(x0;x1)]  find root on interval (f(x0)f(x1)<1)  
rootx[opt;f;(x0;x1)]  root[] with options (as dictionary or
mixed list)
 
solve[eqs;x0]  solve nonlinear equations (given as functions)  
solvex[opt;eqs;x0]  solve[] with options
 
line[f;base;x0]  line search for minimum from base  
linex[opt;f;base;x0]  line[] with same options as rootx[]  
min[f;x0]  find unconstrained minimum  
minx[opt;f;x0]  min[] with same options as solvex[], plus
 
conmin[f;cons;x0]  find constrained minimum (functions cons≥0)  
conminx[opt;f;cons;x0]  min[] with same options as solvex[], plus
