That's weird. It used to find the *.minlexr before the *.gz 

W dniu 2015-02-04 14:42, Tom Hoar napisał(a): 

> I tested more as it is now and the current work-around (delete the extension) 
> has limits. If the reordering-table.minlexr and reordering-table.gz are both 
> present in the referenced folder and there's no extension in the config, 
> moses loads the text .gz model. If you remove the .gz file from the folder, 
> moses finds/loads the .minlexr file. This creates uncertainty and possibly 
> problems for some users.
> 
> I think any/all of your solutions proposals will solve uncertainty. Requiring 
> a fixed extension is probably the easiest. Defining a feature 
> (ReorderingCompact or LexicalReorderingCompact) would make lexical 
> configuration management more like the PhraseDictionary & 
> PhraseDictionaryCompact. There's a lot to be said for consistency.
> 
> Ultimately, I don't have a preference either way. Easy is sometimes better :) 
> For now, we'll use the work-around.
> 
> On 02/04/2015 08:29 PM, Marcin Junczys-Dowmunt wrote: 
> 
> The problem is that Reordering models have a base class that manages and 
> delegates loading. If I remember correctly there is some kind of preference 
> hard coded, if something exists that has the minlexr suffix, take if before 
> *.binphr and *.gz. Some time ago the compact phrase table would not work with 
> the suffix, so this was at least consistent. That changed in the meantime and 
> now we have this current mess. 
> 
> What would be the best solution?
> A) fix suffix handling, 
> B) define a ReorderingCompact feature
> C) both 
> 
> W dniu 2015-02-04 14:18, Tom Hoar napisał(a): Hi, Marcin,
> 
> 1) comment out the reordering model (both, feature and weights): works. 
> Exception disappears.
> 2) Remove file extension ".minlexr" from path= reference: works. file loads 
> and model works
> 
> Additional test:
> 3) Remove file extension ".minphr" from compact phrase table's path= 
> reference: works.
> 
> Is it possible this exception has to do with the filename buffer vice 
> algorithm?
> 
> FYI, I'm using the build-in boost library from Ubuntu 14.04 libboost 1.54
> apt-get install libboost-all-dev
> 
> On 02/04/2015 07:04 PM, Marcin Junczys-Dowmunt wrote: 
> 
> Hi, 
> 
> urgh, I saw this sometimes, could you comment out the reordering model (both, 
> feature and weights) and check if you still get that? If not, uncomment and 
> try to remove the file suffix "minlexr" for the reordering model. If it still 
> happens, update boost :) (someone had the same problem and updating boost 
> fixed it miraculously, no idea why though) 
> 
> W dniu 2015-02-04 12:44, Tom Hoar napisał(a): 
> 
> Hi everyone,
> 
> I just build the new R-3. I'm also in the middle of updating our environment. 
> I had a problem preparing for mert-moses.pl. So, I cut back to basics... 
> manual edits to the moses.ini file and command line.
> 
> I attached the moses.ini file and its output log. I edited the moses.ini file 
> to use compact phrase and reordering files, and KenLM with lazy option. It 
> fails with "Exception: vector::_M_range_check".
> 
> Is something wrong or missing with my edits?
> 
> Also, more results addressed separately in another.
> 
> _______________________________________________
> Moses-support mailing list
> [email protected]
> http://mailman.mit.edu/mailman/listinfo/moses-support [1]
> 
> _______________________________________________
> Moses-support mailing list
> [email protected]
> http://mailman.mit.edu/mailman/listinfo/moses-support [1]

_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support [1]

_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support [1]

_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support [1]

 

Links:
------
[1] http://mailman.mit.edu/mailman/listinfo/moses-support
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to