[hts-users:01225] Re: LBFGS implementation for HTS2.1
- Subject: [hts-users:01225] Re: LBFGS implementation for HTS2.1
- From: "Alexander Gutkin" <alexander.gutkin@xxxxxxxxx>
- Date: Thu, 13 Mar 2008 09:52:50 +0000
- Delivered-to: hts-users@xxxxxxxxxxxxxxx
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:received:received:message-id:date:from:to:subject:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references; bh=kOWXg3G50xL3SrSpFYHiFqQcAneRTLWr28WyoXEaFOU=; b=Y43jkMmP9INwamF31O3vMeYyiD2qxBh8kZWYxBn82y6k42JfToD1qeedwRv6ctq7HJViwc3wiNoq0d8xoD6kDddECUEcqulhkrVITM4Zor4McjsxURVDViQ4uiTNuQcjeloHhQpfZA1C9QV4Ai9OejZQGL+9g20wszByuKTfBJ0=
- Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=message-id:date:from:to:subject:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references; b=e43hz4jgcHvFOB4BeB9zWJJkU6of5RsJTJNFYyaYDHrZb3oanFiblhYMnwPrsJtvzQd2KYDbRX42W2EF4Lm+mU5M1SA1GuIUC9NFw4lcGKBUv68Pr0EBaiQCuXw8w1qfyKDWJKRJKVUfcLCu0lvvZOzamYTcCul7xPfV/SylqHg=
Hi Heiga,
The problem you are experiencing with the g2c port on the 64bit
architecture is most likely due to the wrong byte alignment. If I am
not mistaken, there several places in the code where the control
parameters structure lb3_1 (accuracy etc.) is accessed through
dereferencing of lb3. That is not necessarily guaranteed to work.
Regards,
Alexander.
On Thu, Mar 13, 2008 at 2:10 AM, Heiga ZEN (Byung Ha CHUN)
<zen@xxxxxxxxxxxxxxx> wrote:
> Hi,
>
> Sacha Krstulovic wrote (2008/03/13 1:55):
>
>
> > For the HTS2.1 release, is the LBFGS implementation going to remain
> > written in Fortran, or will it be replaced by a C code implementation?
>
> That LBFGS FORTRAN code originally comes from Jorge Nocedal, professor of Northwestern University.
> I heard that it has been a public domain, so we can modify and redistribute it.
> Actually I have one in C ported by g2c.
> However, it has some problem on x86_64 environment, so I don't use it for now.
>
>
> > In hts_engine from HTS-API 0.95, does the GV-related optimization
> > correspond to the LBFGS algorithm, or is the C-code implementing
> > a different optimization algorithm?
>
> LBFGS is computationally more expensive than the steepest descent and quasi-Newton methods.
> We design HMGenS for research and hts_engine API is for application.
> Therefore, in hts_engine API, the quasi-Newton method (diagonal Hessian matrix) is used.
> You can find the codes for this optimization at HTS_calc_derivative() and HTS_gv_parmgen() in lib/HTS_mlpg.c.
>
> Regards,
>
> Heiga ZEN (Byung Ha CHUN)
>
> --
> ------------------------------------------------
> Heiga ZEN (in Japanese pronunciation)
> Byung Ha CHUN (in Korean pronunciation)
>
> Department of Computer Science and Engineering
> Nagoya Institute of Technology
> Gokiso-cho, Showa-ku, Nagoya 466-8555 Japan
>
> http://www.sp.nitech.ac.jp/~zen
> ------------------------------------------------
>
>
- References
-
- [hts-users:01219] LBFGS implementation for HTS2.1, Sacha Krstulovic
- [hts-users:01221] Re: LBFGS implementation for HTS2.1, Heiga ZEN (Byung Ha CHUN)