You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@opennlp.apache.org by Jörn Kottmann <ko...@gmail.com> on 2013/09/04 11:59:00 UTC
Re: Question about the zipped model files
On 08/26/2013 03:00 PM, Ioan Barbulescu wrote:
> Hi guys
>
> Short question, please:
>
> Currently, the opennlp models come as zipped files.
>
> Is it possible to use them in an expanded / un-zipped form?
> (and how?)
>
> Zipped is very neat and clean, but it adds some time when reading the file.
> I am interested in speeding up as much as possible the load time.
You can probably repackage the zip files without using compression.
Anyway I doubt that it adds much time, did you profile the loading code?
As far as I know is the slowest part to build the maxent model, maybe that
can be speed up, I never profiled that part of OpenNLP.
The life-cycle of a model should be the same as of your application,
maybe you can
just find a way to reuse them, instead of loading them over and over again.
Jörn
Re: Question about the zipped model files
Posted by Aliaksandr Autayeu <al...@autayeu.com>.
RAM Drive with models?
On Sat, Sep 7, 2013 at 8:47 PM, Ioan Barbulescu <ib...@gmail.com>wrote:
> Hi guys
>
> Thank you for your answers.
>
> Just to clarify the reasons for my question:
>
> - I am at the stage where I write my code, so I have to run my unit tests
> like 50 times a day :)
> Hence, every second shaven off from the start time is invaluable.
> Yes, I agree with Jörn, the lifecycle of the model is indeed the same
> with the one of my application.
>
> - I am already ussing a decent SSD, so loading time as a factor is
> off-the-table.
>
> I will keep tinker around it and let you know how it goes ...
>
> Many thanks.
> Ioan
>
> On Sat, Sep 7, 2013 at 9:21 PM, Aliaksandr Autayeu
> <al...@autayeu.com>wrote:
>
> > Ioan, often hard drive speed limits you more than processor speed. That
> is,
> > it might be faster to load 5M from disk and unpack than load unpacked
> > 25M from disk.
> >
> > Aliaksandr
> >
> >
> > On Wed, Sep 4, 2013 at 11:59 AM, Jörn Kottmann <ko...@gmail.com>
> wrote:
> >
> > > On 08/26/2013 03:00 PM, Ioan Barbulescu wrote:
> > >
> > >> Hi guys
> > >>
> > >> Short question, please:
> > >>
> > >> Currently, the opennlp models come as zipped files.
> > >>
> > >> Is it possible to use them in an expanded / un-zipped form?
> > >> (and how?)
> > >>
> > >> Zipped is very neat and clean, but it adds some time when reading the
> > >> file.
> > >> I am interested in speeding up as much as possible the load time.
> > >>
> > >
> > > You can probably repackage the zip files without using compression.
> > > Anyway I doubt that it adds much time, did you profile the loading
> code?
> > >
> > > As far as I know is the slowest part to build the maxent model, maybe
> > that
> > > can be speed up, I never profiled that part of OpenNLP.
> > >
> > > The life-cycle of a model should be the same as of your application,
> > > maybe you can
> > > just find a way to reuse them, instead of loading them over and over
> > again.
> > >
> > > Jörn
> > >
> > >
> > >
> >
>
Re: Question about the zipped model files
Posted by Ioan Barbulescu <ib...@gmail.com>.
Hi guys
Thank you for your answers.
Just to clarify the reasons for my question:
- I am at the stage where I write my code, so I have to run my unit tests
like 50 times a day :)
Hence, every second shaven off from the start time is invaluable.
Yes, I agree with Jörn, the lifecycle of the model is indeed the same
with the one of my application.
- I am already ussing a decent SSD, so loading time as a factor is
off-the-table.
I will keep tinker around it and let you know how it goes ...
Many thanks.
Ioan
On Sat, Sep 7, 2013 at 9:21 PM, Aliaksandr Autayeu
<al...@autayeu.com>wrote:
> Ioan, often hard drive speed limits you more than processor speed. That is,
> it might be faster to load 5M from disk and unpack than load unpacked
> 25M from disk.
>
> Aliaksandr
>
>
> On Wed, Sep 4, 2013 at 11:59 AM, Jörn Kottmann <ko...@gmail.com> wrote:
>
> > On 08/26/2013 03:00 PM, Ioan Barbulescu wrote:
> >
> >> Hi guys
> >>
> >> Short question, please:
> >>
> >> Currently, the opennlp models come as zipped files.
> >>
> >> Is it possible to use them in an expanded / un-zipped form?
> >> (and how?)
> >>
> >> Zipped is very neat and clean, but it adds some time when reading the
> >> file.
> >> I am interested in speeding up as much as possible the load time.
> >>
> >
> > You can probably repackage the zip files without using compression.
> > Anyway I doubt that it adds much time, did you profile the loading code?
> >
> > As far as I know is the slowest part to build the maxent model, maybe
> that
> > can be speed up, I never profiled that part of OpenNLP.
> >
> > The life-cycle of a model should be the same as of your application,
> > maybe you can
> > just find a way to reuse them, instead of loading them over and over
> again.
> >
> > Jörn
> >
> >
> >
>
Re: Question about the zipped model files
Posted by Aliaksandr Autayeu <al...@autayeu.com>.
Ioan, often hard drive speed limits you more than processor speed. That is,
it might be faster to load 5M from disk and unpack than load unpacked
25M from disk.
Aliaksandr
On Wed, Sep 4, 2013 at 11:59 AM, Jörn Kottmann <ko...@gmail.com> wrote:
> On 08/26/2013 03:00 PM, Ioan Barbulescu wrote:
>
>> Hi guys
>>
>> Short question, please:
>>
>> Currently, the opennlp models come as zipped files.
>>
>> Is it possible to use them in an expanded / un-zipped form?
>> (and how?)
>>
>> Zipped is very neat and clean, but it adds some time when reading the
>> file.
>> I am interested in speeding up as much as possible the load time.
>>
>
> You can probably repackage the zip files without using compression.
> Anyway I doubt that it adds much time, did you profile the loading code?
>
> As far as I know is the slowest part to build the maxent model, maybe that
> can be speed up, I never profiled that part of OpenNLP.
>
> The life-cycle of a model should be the same as of your application,
> maybe you can
> just find a way to reuse them, instead of loading them over and over again.
>
> Jörn
>
>
>