[developers] Fw: ERG installation questions

Tue Mar 26 05:35:49 CET 2019

Hi Neil,

I can address some of your points.

On Tue, Mar 26, 2019 at 6:00 AM Neil Stoker <neil at nmstoker.com> wrote:

> Thanks Dan. That's really helpful and somehow I'd missed seeing that Quick
> Start page, so I'll study it to see what I can pick up.
>
> On Mon, 25 Mar 2019, 21:05 Dan Flickinger, <danf at stanford.edu> wrote:
>
>> Developers and Neil,
>>
>>
>> I'm forwarding to our more active `developers' list a request below from
>> Neil that I hope others of you can answer with more clarity than I have.
>> Neil, at the risk of presenting too many options, you might also look at
>> this wiki page which outlines an alternative setup using the ACE parser and
>> a suite of Python tools called PyDelphin:
>>
>>
>> http://moin.delph-in.net/QuickStart
>>
>>
>> Thanks for your interest!
>>
>>  Dan
>>
>>
>> ------------------------------
>> *From:* Neil Stoker <neil at nmstoker.com>
>> *Sent:* Sunday, March 24, 2019 6:06 PM
>> *To:* lingo at delph-in.net
>> *Subject:* ERG installation questions
>>
>> Hello,
>>
>> I had a couple of questions about the installation of ERG and wondered if
>> you could help?
>>
>> I've been looking over the various pages detailing installation and
>> background info (eg as per links below) as I'm hoping to install ERG on my
>> main computer at home which runs Arch Linux and I wanted to check a couple
>> of things first. I'm currently expecting to go for the compete install (ie
>> option A)
>>
>> http://moin.delph-in.net/ErgProcessing
>> http://moin.delph-in.net/LogonInstallation
>>
>> 1. Has there been any exploration of using Docker to simplify
>> installation? It seems to me that a lot of the complexity could be isolated
>> if there were a Docker script. I'm not a Docker expert but I might try to
>> have a go with this approach, assuming I'm not reinventing the wheel here!
>>
>> 2. There is mention of 32- bit binaries, but it also says that's as of
>> 2012. Given the trend toward 64-bit in the meantime, I wondered if that had
>> now changed?
>>
>> If not, is the source for those binaries available or is the point that
>> they're only distributed pre-compiled (ie 32-bit)?
>>
>
32-bit libraries appear to still be required for certain functionality. On
Pop!_OS 18.10 (an Ubuntu derivative), after downloading the entire LOGON
tree I was able to start up the LKB but not [incr tsdb()]. There were
errors about 'yzlui' and 'swish++' missing, but the error was misleading:
the binaries were at the specified location but the necessary 32-bit shared
libraries were not installed. By installing these according to
http://moin.delph-in.net/LogonUbuntu (just the first block about "recent"
Ubuntus) I was able to make the errors go away and [incr tsdb()] could
load. Alternatively, I could get a 64-bit yzlui from
http://sweaglesw.org/linguistics/yzlui.x86-64 and replace the one at
$LOGONROOT/logon/lingo/lkb/bin/linux.x86.64/yzlui, but I could not find a
64-bit swish++ binary. Also see this thread for more info about yzlui:
http://lists.delph-in.net/archives/developers/2018/002872.html.

Also, I used to use Arch Linux and at the time I wrote some instructions
for installing LOGON: http://moin.delph-in.net/LogonArch. The last edit was
2013, so it may be quite outdated, but perhaps you'll find it helpful.

> 3. Is it strictly x86 or could this be made to work on ARM architectures?
>> (guess that would depend on the response re the 32-bit point above).
>>
>> 4. Is there a way to expose the REST API from a local installation of
>> ERG? I'm likely to want to run it on a large amount of data and didn't want
>> to abuse the Oslo facilities, but using the API locally was appealing.
>>
>
I'm not aware of any instructions to run LOGON's web API locally, but the
Bottlenose server (https://github.com/delph-in/bottlenose) is easy to set
up. It is built on top of PyDelphin and ACE, so you may get slightly
different results than from the LOGON version, but both implement the same
API so it would be easy to switch from one to the other. In fact, the
Delphin-viz demo uses both servers in the same installation:
http://delph-in.github.io/delphin-viz/demo

However if you wish to run the ERG on large amounts of data, I'd look into
processing testsuites (also called "profiles"). Doing so may be more
efficient than running parses through the web API and also you'll get more
diagnostic information such as memory or time usage, etc., which can be
useful for comparisons or debugging. The following software can process
testsuites:

* [incr tsdb()] -- http://moin.delph-in.net/ItsdbTop
* art -- http://sweaglesw.org/linguistics/libtsdb/art
* PyDelphin --
https://pydelphin.readthedocs.io/en/latest/tutorials/itsdb.html#processing-testsuites-with-ace

> I hope that's not too much to ask at once! A big thank you in advance, as
>> this looks like a great project and I think it will be really interesting
>> for analysis I'm working on.
>>
>> Kind regards,
>> Neil Stoker
>>
>

-- 
-Michael Wayne Goodman
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.delph-in.net/archives/developers/attachments/20190326/130b881e/attachment-0001.html>