Re: [Kaldi-users] FST minimization of LMs

[Daniel P. <dp...@gm...>]

No, because for each path there will be multiple alternatives in B and the
sum of the path weights will be more than one.  I recommend to address
further questions on this topic to www.openfst.org.
Dan



On Thu, Dec 19, 2013 at 8:53 PM, Lahiru Samarakoon <lah...@gm...>wrote:

> Is this also applies to the log semiring?
> Thanks
> Lahiru
>
>
> On Fri, Dec 20, 2013 at 12:46 PM, Daniel Povey <dp...@gm...> wrote:
>
>>
>> I am trying to decompose weighted acceptors (Language models) . I observe
>>> this behavior when I compose them to build a new fst. I think I am doing
>>> something wrong here.
>>>
>>> Please be kind enough to clarify me the following.
>>>
>>> Given a *weighted acceptor *"*A*",  lets say acceptor "*B*" is created
>>> by setting all the weights of "A" to *zero *including the final states
>>> weights.
>>> is
>>> *"A compose B"  equivalent to "A"?*
>>>
>>
>> If you are in the tropical semiring, then yes, because any path in A is
>> also in B (with zero cost).  But "A compose B" is a wasteful way of
>> representing the acceptor because there are multiple alternative paths
>> through B that will end up multiplying the size, potentially by the n-gram
>> order minus one.  So it's equivalent in the FST sense, but it's much
>> larger.
>>
>> Dan
>>
>>
>>
>>>
>>> if the answer is not  equivalent, could you state the reason briefly
>>>
>>> Thank you,
>>>
>>> Best Regards,
>>> Lahiru
>>>
>>>
>>> On Fri, Dec 20, 2013 at 3:41 AM, Al Zatv <al...@gm...> wrote:
>>>
>>>> Hello! Actually, I don't have such a problem. I think Lahiru
>>>> Samarakoon, the topic starter, have it:
>>>> Quote:
>>>> =====
>>>>
>>>> I have my own language model. When I minimize it, actually it increases
>>>> the number of states and arcs.
>>>>
>>>> How can this be possible?
>>>> I have created the LM in arpa format and used scripts from Kaldi to
>>>> create the G.fst
>>>> =====
>>>>
>>>> Usually ARPA LMs are acceptors and they are determinized by default, so
>>>> if Lahiru really observe that min(G.fst) is bigger than G.fst, it is
>>>> suspicious.
>>>> 19.12.2013 20:09 пользователь "Michael Riley" <ri...@go...>
>>>> написал:
>>>>
>>>> Not sure what exactly you are doing or what code you are calling but if
>>>>> you are using the algorithms in OpenFst then
>>>>> |det(A)| can be > that |A| but |min(det(A))| <= |det(A)| if A is an
>>>>> acceptor.
>>>>>
>>>>> I'd be surprised if there is a bug here (given this condition I
>>>>> believe is tested on millions of randomly generated machines) but If you
>>>>> have a counter-example when using unmodified OpenFst as above, then send it
>>>>> to us along with the precise commands you ran that demonstrate the problem.
>>>>>
>>>>> -m
>>>>>
>>>>> -m
>>>>>
>>>>>
>>>>> On Thu, Dec 19, 2013 at 2:55 AM, Al Zatv <al...@gm...> wrote:
>>>>>
>>>>>> According to this document, size of transducer can be larger than
>>>>>> original. But ARPA LM is an acceptor, so it must have optimal number of
>>>>>> states, or I don't misunderstood something?
>>>>>> 19.12.2013 10:23 пользователь "Michael Riley" <ri...@go...>
>>>>>> написал:
>>>>>>
>>>>>> http://www.openfst.org/twiki/bin/view/FST/MinimizeDoc
>>>>>>>
>>>>>>>
>>>>>>> On Thu, Dec 19, 2013 at 1:13 AM, Lahiru Samarakoon <
>>>>>>> lah...@gm...> wrote:
>>>>>>>
>>>>>>>>
>>>>>>>> HI All,
>>>>>>>>
>>>>>>>> I have my own language model. When I minimize it, actually it
>>>>>>>> increases the number of states and arcs.
>>>>>>>>
>>>>>>>> How can this be possible?
>>>>>>>> I have created the LM in arpa format and used scripts from Kaldi to
>>>>>>>> create the G.fst.
>>>>>>>>
>>>>>>>> Thank you,
>>>>>>>>
>>>>>>>> Best Regards,
>>>>>>>> Lahiru
>>>>>>>>
>>>>>>>>
>>>>>>>> ------------------------------------------------------------------------------
>>>>>>>> Rapidly troubleshoot problems before they affect your business.
>>>>>>>> Most IT
>>>>>>>> organizations don't have a clear picture of how application
>>>>>>>> performance
>>>>>>>> affects their revenue. With AppDynamics, you get 100% visibility
>>>>>>>> into your
>>>>>>>> Java,.NET, & PHP application. Start your 15-day FREE TRIAL of
>>>>>>>> AppDynamics Pro!
>>>>>>>>
>>>>>>>> http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk
>>>>>>>> _______________________________________________
>>>>>>>> Kaldi-users mailing list
>>>>>>>> Kal...@li...
>>>>>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> ------------------------------------------------------------------------------
>>>>>>> Rapidly troubleshoot problems before they affect your business. Most
>>>>>>> IT
>>>>>>> organizations don't have a clear picture of how application
>>>>>>> performance
>>>>>>> affects their revenue. With AppDynamics, you get 100% visibility
>>>>>>> into your
>>>>>>> Java,.NET, & PHP application. Start your 15-day FREE TRIAL of
>>>>>>> AppDynamics Pro!
>>>>>>>
>>>>>>> http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk
>>>>>>> _______________________________________________
>>>>>>> Kaldi-users mailing list
>>>>>>> Kal...@li...
>>>>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users
>>>>>>>
>>>>>>>
>>>>>
>>>
>>>
>>> ------------------------------------------------------------------------------
>>> Rapidly troubleshoot problems before they affect your business. Most IT
>>> organizations don't have a clear picture of how application performance
>>> affects their revenue. With AppDynamics, you get 100% visibility into
>>> your
>>> Java,.NET, & PHP application. Start your 15-day FREE TRIAL of
>>> AppDynamics Pro!
>>>
>>> http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk
>>> _______________________________________________
>>> Kaldi-users mailing list
>>> Kal...@li...
>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users
>>>
>>>
>>
>