Moses-support Digest, Vol 91, Issue 37

Send Moses-support mailing list submissions to
moses-support@mit.edu

To subscribe or unsubscribe via the World Wide Web, visit
http://mailman.mit.edu/mailman/listinfo/moses-support
or, via email, send a message with subject or body 'help' to
moses-support-request@mit.edu

You can reach the person managing the list at
moses-support-owner@mit.edu

When replying, please edit your Subject line so it is more specific
than "Re: Contents of Moses-support digest..."


Today's Topics:

1. Re: Tuning crashed (Wei Qiu)


----------------------------------------------------------------------

Message: 1
Date: Sat, 24 May 2014 14:57:59 +0200
From: Wei Qiu <wei@qiu.es>
Subject: Re: [Moses-support] Tuning crashed
To: Barry Haddow <bhaddow@staffmail.ed.ac.uk>
Cc: moses-support <moses-support@mit.edu>
Message-ID:
<CALWr_T-=tLox7D6Btmz09=3+yNzNLa5a1oqF3SEWL4Mu7mUgHw@mail.gmail.com>
Content-Type: text/plain; charset="utf-8"

Thanks, Barry,

Both batches-mira and pro tuning don't crash. But the result of
batch-miratuning is very unstable due to the size of dev set.

Best,
Wei



On Sat, May 24, 2014 at 11:34 AM, Wei Qiu <wei@qiu.es> wrote:

> Thanks, Barry. I will try mira.
>
> Anyway, I share my model file here
>
> https://www.dropbox.com/s/jko7aitnk9mqa8a/model.tar.gz
>
> Best,
> Wei
>
>
> On Sat, May 24, 2014 at 10:13 AM, Barry Haddow <bhaddow@staffmail.ed.ac.uk
> > wrote:
>
>> Hi Wei
>>
>> This error (as far as I recall) tends to mean that one of your features
>> is flat (ie uninformative). I don't see that this is the case for your
>> features. The only thing I could think of is that your tuning set is quite
>> small (500 sentences) but that doesn't usually cause a problem.
>>
>> My suggestion would be to try tuning with batch mira or pro.
>>
>> cheers - Barry
>>
>>
>> On 24/05/14 08:31, Wei Qiu wrote:
>>
>> Hi,
>>
>> I tried to train a phrase-based baseline system from German to English.
>>
>> But during the tuning step, mert crashed in the first run. I tracked it
>> to this error,
>>
>> Check abs(leftmost->first-gradient.rbegin()->first) < 0.0001 failed in
>> mert/Optimizer.cpp:170
>>
>> I've already checked similar posts on the mailing list, but my run1. outand run1. Scores
>> .data look alright.
>>
>> The tuning directory is shared here:
>> https://www.dropbox.com/s/j6tflbtft3uqvuu/tmp.1.tar.gz
>>
>> The model directory will be shared as well(it's still uploading).
>>
>> Thanks in advance.
>>
>> Best,
>> Wei Qiu
>>
>>
>>
>>
>> _______________________________________________
>> Moses-support mailing listMoses-support@mit.eduhttp://mailman.mit.edu/mailman/listinfo/moses-support
>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/mailman/private/moses-support/attachments/20140524/dcff41f7/attachment-0001.htm

------------------------------

_______________________________________________
Moses-support mailing list
Moses-support@mit.edu
http://mailman.mit.edu/mailman/listinfo/moses-support


End of Moses-support Digest, Vol 91, Issue 37
*********************************************

Tweet