cl-libsvm

CL-LIBSVM

Common Lisp wrapper for the libsvm support vector machine library by
Ravenpack, Gabor Melis.

It's under the MIT licence. See COPYING.

* Design notes

The lisp API tries to protect the programmer from the dangers of C,
take care of memory management and wrap everything in CLOS classes
while trying not to deviate too much from the original interface so
that one can use existing documentation without too much trouble.

* API differences

PROBLEMs, PARAMETERs and MODELs are represented as CLOS classes.
Foreign memory is freed when they are garbage collected. The
implementation is based on weak value hash tables (currently for sbcl,
allegro and clisp).

In the C API, one passes sparse vectors as struct node *, where a node
with index -1 denotes the end of the vector. On the lisp side one
passes sparse vectors as vectors of index/value conses like (VECTOR
(CONS 2 0.1) (CONS 3 0.2)) or as mapper functions. The restrictions
that indices should be increasing and not less than 1 still stand, but
unlike in C, violations are detected.

A mapper for a sparse vector calls the function it is given with index
and value parameters. A mapper function that simply maps a vector of
conses is:

  (lambda (fn)
    (map nil (lambda (c)
               (funcall fn (car c) (cdr c)))
         (vector (cons 2 0.1) (cons 3 0.2))))

Mappers abstract away the underlying data structure. Generalized
sequences would be nicer, but they only exist in SBCL.

One can pass a mapper function instead of a vector anywhere including
the TARGETS parameter of TRAIN. It also works for the INPUTS parameter
since INPUTS is a vector of sparse vectors one can turn it into a
mapper to sparse vector mappers.

* Compatibility

Tested with libsvm version 2.82, 2.83, 2.86 and 2.88.

Some functionality (MODEL-W2S) requires patching the original sources.
The distributed sources under lib/src/ contain the necessary changes.

In the lib/ directory there are platform specific shared libraries,
see the README files for version information.

Note that you may want to disable excessive verbosity by changing '#if
1' to '#if 0' around the definition of info in svm.cpp. The
precompiled libraries in lib/ contain this change.


cffi

trivial-garbage

CFFI wrapper for LIBLINEAR, the machine learning library

cl-liblinear

CFFI wrapper for LIBSVM, the machine learning library

cl-libsvm

Upstream URL

Author

License

cl-liblinear

Version:

License:

Authors:

Dependencies:

cl-libsvm

Version:

License:

Authors:

Dependencies:

Dependencies (2)

Dependents (0)