From mboxrd@z Thu Jan  1 00:00:00 1970
Received: from mail1-relais-roc.national.inria.fr (mail1-relais-roc.national.inria.fr [192.134.164.82])
	by walapai.inria.fr (8.13.6/8.13.6) with ESMTP id p3KE0vab025082
	for <caml-list@sympa-roc.inria.fr>; Wed, 20 Apr 2011 16:00:57 +0200
X-IronPort-Anti-Spam-Filtered: true
X-IronPort-Anti-Spam-Result: ApACANzlrk3RVaG2kWdsb2JhbACEUJMhjUMIFAEBAQEJCwsHFAQhiG+jKYo4PIIlhTcziF0BAQMGgSODTnoEjiKDf4EEhH86
X-IronPort-AV: E=Sophos;i="4.64,246,1301868000"; 
   d="scan'208";a="106332391"
Received: from mail-gx0-f182.google.com ([209.85.161.182])
  by mail1-smtp-roc.national.inria.fr with ESMTP/TLS/RC4-SHA; 20 Apr 2011 16:00:51 +0200
Received: by gxk28 with SMTP id 28so304889gxk.27
        for <caml-list@inria.fr>; Wed, 20 Apr 2011 07:00:50 -0700 (PDT)
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed;
        d=gmail.com; s=gamma;
        h=domainkey-signature:message-id:date:from:user-agent:mime-version:to
         :subject:references:in-reply-to:content-type
         :content-transfer-encoding;
        bh=eUFZHA20OKzXXYtqK+sqBZ0KjlsVCOv6S6xzGri1dFo=;
        b=KHeBZyFgHlSDgMe2Aefc61ksCnwNldxhO28Nt3UVO2hfBJf6amqudjFBRD37ec9oGo
         d8G5zKkwZSkDIKIO4L4PYbGLDCTynOGcCz0cnW+SJkyvje1NYRnAo25e7OpD+mquCX++
         bFYiV9MRQPom+Cj3NGna6n5PwdAptbiqRei8E=
DomainKey-Signature: a=rsa-sha1; c=nofws;
        d=gmail.com; s=gamma;
        h=message-id:date:from:user-agent:mime-version:to:subject:references
         :in-reply-to:content-type:content-transfer-encoding;
        b=CRJLX7kY2oigfgsdSAzgZw2FQ+6WvMhvf0mbrYyZy8bguGLnpFpP8wresuNoV+WWjg
         DdhX9YY/c6XfFkGCp+/HFBvAmyP+uwz+9pgJPk8bhXFN8OWPU3nNGJv2bgWCB5jngKau
         BJsu+zW1v4+7z4GY3jIyLhtIMC8QldJ2vUOiU=
Received: by 10.90.6.13 with SMTP id 13mr6269030agf.75.1303308050659;
        Wed, 20 Apr 2011 07:00:50 -0700 (PDT)
Received: from [192.168.1.65] (99-121-78-10.lightspeed.lnngmi.sbcglobal.net [99.121.78.10])
        by mx.google.com with ESMTPS id l14sm904344anc.50.2011.04.20.07.00.47
        (version=SSLv3 cipher=OTHER);
        Wed, 20 Apr 2011 07:00:48 -0700 (PDT)
Message-ID: <4DAEE70E.4060401@gmail.com>
Date: Wed, 20 Apr 2011 10:00:46 -0400
From: Edgar Friendly <thelema314@gmail.com>
User-Agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv:1.9.2.14) Gecko/20110223 Thunderbird/3.1.8
MIME-Version: 1.0
To: caml-list@inria.fr
References: <2054357367.219171.1300974318806.JavaMail.root@zmbs4.inria.fr>	 <4D8BD02D.1010505@inria.fr> <4D8C73C8.6020801@inescporto.pt>	 <1301055903.8429.314.camel@thinkpad>	 <341494683.237537.1301057887481.JavaMail.root@zmbs4.inria.fr>	 <4D8C944A.9060601@inria.fr> <4D8CB859.9040709@inescporto.pt>	 <4D8CDDCC.4010000@ens-lyon.org> <4D8CEAA4.2030403@inescporto.pt>	 <BANLkTikKSVLZcffTn6Ku8eZUyDdDT8cykA@mail.gmail.com> <1303244809.8429.1272.camel@thinkpad> <4DAE9278.4050701@inescporto.pt>
In-Reply-To: <4DAE9278.4050701@inescporto.pt>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Subject: Re: [Caml-list] Efficient OCaml multicore -- roadmap?


>> Hugo, I'd like to do some more experiments into this direction. Is there
>> a simple version of machine learning algorithm I could try to
>> parallelize?
>>
>
If you're interested in parallelizing Machine Learning algorithms, I've 
had to write a number of them for a class project, and I did it from 
scratch in ocaml (just using BLAS primitives for some fast vector math.  
They're here: https://github.com/thelema/machine-learning  Small parts 
of what I'm doing depend on libosvm, my fork of libsvm that's not 
published yet (just build fixes, needs more work on the types), but 
that's not the part you're interested in, so if you want to compile it, 
just remove the svm_* calls, and it should compile just fine.

The algorithms are all in predict.ml, grouped into binary classifiers 
(which are generally online = sequential, except for klsq (kernel least 
squares) which is just a matrix inversion at its core) and extension 
routines that take a binary classifier and run it many times on the data 
in different ways to get a multi-way classifier.  This second group is 
*ridiculously* parallelizable.

There's also routines for running the produced classifiers on test data, 
which parallelize across different elements of the test data independently.

I wish my deadline weren't today, as I'd love to have a parallel version 
of this code to run my huge dataset through.
E.