Re: LARGE SCALE MACHINE LEARNING SUPPORT IN LISP WORKS
Hi,
No I don't use CL for matrix libraries, I've written what there is no sense to
implement matrix library in CL because there are tons of them really
well optimized and well tested in C, and one just have to write good
wrappers for them(believe me it is still easier than implement the good
matrix library). I understand those who implements
these libraries in CL, but personaly I wouldn't do that.
Robert Smith <quad@symbo1ics.com> writes:
> Alexey:
>
> Making wrappers for them is a very non-trivial thing. Do you use Lisp
> arrays? Foreign arrays?
>
> If you use Lisp arrays, what do you do with GC? Pinning (non-portable,
> non-reliable)? Stopping GC (possible concurrency issues)?
>
> If you use foreign arrays, do you re-write all of Lisp's access
> functions? How do you interface with Lisp array libraries? How do you
> deal with GC? Finalizers (non-reliable)? What if you run out of
> foreign memory space? How do you detect that and do a GC so the
> finalizers can kick off?
>
> Writing a proof-of-concept wrapper is very easy, but Getting It Right
> (TM) is very difficult, in my opinion.
>
> (FWIW: Matlisp, as conceived by Raymond Toy, used Lisp arrays with
> stop-the-GC approach.)
>
> Robert
>
> On Mon, Sep 26, 2016 at 12:16 PM, Alexey Veretennikov
> <alexey.veretennikov@gmail.com> wrote:
>> Linear algebra is not a problem in C with tons of high-quality and
>> highly optimized
>> commercial and open-source libraries available(BLAS/uBLAS, LAPACK,
>> Intell's MKL, Armadillo, Matrix Template Library, SuiteSparse to name a
>> few), with invested many man-years and top-of-the-edge research behind them,
>> so I guess better just to have good wrappers for them.
>>
>> Robert Smith <quad@symbo1ics.com> writes:
>>
>>> Linear algebra in Common Lisp is currently not at all a solved problem.
>>>
>>> Every library that I know of, except a proprietary one I used, either has an
>>> awful interface, is very slow,
>>> or uses the most insane and arcane Lisp MOP+macrology ever.
>>>
>>> This is a space that definitely needs to be filled. I have ideas on how to
>>> approach it, but time is limited. If
>>> anyone is interested in helping design/build such a library, do get in touch.
>>>
>>> Robert
>>>
>>> On Tue, Sep 20, 2016 at 2:10 PM, Laughing Water <lw@mt.net> wrote:
>>>
>>>
>>> Check this out:
>>>
>>> http://www.cliki.net/linear%20algebra
>>>
>>> Among other links on the page, there’s cl-octave, which is a foreign
>>> function interface to the Octave
>>> language that’s used in your (our) course, and there’s MGL-MAT that supports
>>> interfacing to CUDA
>>> code.
>>>
>>> I’m thankful you asked that question. Good luck with the course!
>>>
>>> LW
>>>
>>> On Sep 20, 2016, at 6:53 AM, Seyi Akadri <seyiakadri@gmail.com> wrote:
>>>
>>> Hello,
>>> I am currently learning machine learning with Prof. Andrew of Stanford
>>> university via an online
>>> course in coursera.
>>>
>>> The implementations for the ML course is mostly in Matlab/Octave.
>>>
>>> I was wondering if i can get a good book on large scale machine learning in
>>> common Lisp.
>>>
>>> Although i have the practical artificial intelligence in Lisp book, i have
>>> not not really gone
>>> through the book.
>>>
>>> I know this is a LW forum, but i won't mind if i can get some pointers in
>>> the area of ML using
>>> Lisp, especially for large scale machine learning using large training
>>> examples, big data and
>>> perhaps multi-core processing.
>>>
>>> Does LW support this? I mean for large scale ML involving big data and
>>> multi-core processing.
>>>
>>> Thank you very much
>>> Seyi Akadri
>>>
>>>
>>>
>>
>> --
>> Br,
>> /Alexey
--
Br,
/Alexey
_______________________________________________
Lisp Hug - the mailing list for LispWorks users
lisp-hug@lispworks.com
http://www.lispworks.com/support/lisp-hug.html