[Cbc] Cbc Digest, Vol 92, Issue 5

[John Forrest]

Yves,

I am not sure in what sense you are using "deterministic".

I assumed you meant in a loose sense so that it annoyed you that adding 
an heuristic which did not work changed the results. Obviously adding an 
heuristic that does work will change results. Making any change may make 
a change:-)

If you mean truly deterministic i.e. two runs of same model with same 
parameters give different results, then that is a bug - can you give me 
an example?

setRandomSeed(0) generates a seed depending on clock - -1 is deterministic.

I have modified code so that -DSAME_HEURISTIC_SEED switches off that 
particular variability (I did not think it of enough general use to make 
it a -- switch).  I think that variability can be useful e.g. two copies 
of same heuristic but with a different seed.

John

On 08/02/15 08:42, Yves Touchard wrote:
> John,
>
> Determinism is mandatory for our process.
> So, any way to control the seed will be appreciate!
> Maybe, you could also implement this feature with a command-line 
> argument (--randomHeuristicSeed)?
>
> Two other questions:
>
>   * What is the exact use of randomCbcSeed. I don't set it (and did
>     not notice and determinism problem) and the default value (-1)
>     seems to generate a new seed each time cbc is launched?
>   * Setting cutlength to 10000030 leads to determinism problems. Is it
>     also expected?
>
> Thanks.
>
> Yves
>
>
> Le 07/02/2015 18:00, cbc-request at list.coin-or.org a écrit :
>> Send Cbc mailing list submissions to
>> 	cbc at list.coin-or.org
>>
>> To subscribe or unsubscribe via the World Wide Web, visit
>> 	http://list.coin-or.org/mailman/listinfo/cbc
>> or, via email, send a message with subject or body 'help' to
>> 	cbc-request at list.coin-or.org
>>
>> You can reach the person managing the list at
>> 	cbc-owner at list.coin-or.org
>>
>> When replying, please edit your Subject line so it is more specific
>> than "Re: Contents of Cbc digest..."
>>
>>
>> Today's Topics:
>>
>>     1. Re: Determinism with divingV option (John Forrest)
>>
>>
>> ----------------------------------------------------------------------
>>
>> Message: 1
>> Date: Fri, 06 Feb 2015 17:15:56 +0000
>> From: John Forrest<john.forrest at fastercoin.com>
>> To:cbc at list.coin-or.org
>> Subject: Re: [Cbc] Determinism with divingV option
>> Message-ID:<54D4F6CC.2060404 at fastercoin.com>
>> Content-Type: text/plain; charset="windows-1252"; Format="flowed"
>>
>> Yves,
>>
>> It is to be expected!  When an heuristic is added it is given an initial
>> seed for use in generating pseudo random numbers.  This seed is changed
>> depending on k for the k'th heuristic.  So adding divingVectorLength
>> changes the seed for the RINS heusristic.
>>
>> If you don't like that, I can easily add an ifdef so
>> -DSAME_HEURISTIC_SEED will switch off - in CbcModel:addHeuristic.
>>
>> John
>>
>>
>>
>> On 06/02/15 14:03, Yves Touchard wrote:
>>> Hello,
>>>
>>> I got a determinism problem with the lp file (and maybe others)
>>> located here: /https://www.dropbox.com/s/0i7c84lbm2pvyl3/t1.lp?dl=0/.
>>> The solution file is different  whether the divingV option is set or not.
>>> Command-line is the following:
>>>
>>> /b//in/cbc t1.lp threads 104 ratio 0.05 cutlength 30 *divingV* on
>>> divesolves 301 multiple 1 branch printingOptions rows solution t1_0.sol/
>>>
>>> Solution files are located here:
>>>
>>>    * Without -> /https://www.dropbox.com/s/zcu8kwv9ukipaqe/t1_0.sol?dl=0/
>>>    * With -> /https://www.dropbox.com/s/jc1co4yeg9i13vw/t1_1.sol?dl=0/
>>>
>>> I tried with both trunk revision 2102 and 2128.
>>> Could you tell if it is an expected behaviour or not?
>>>
>>> Thanks & Regards
>>>
>>> Yves
>>>
>>> ------------------------------------------------------------------------------
>>> This message and any attachments (the "message") are confidential and
>>> intended solely for the addressee(s). Any unauthorized use or
>>> dissemination is prohibited. E-mails are susceptible to alteration.
>>> Neither DxO Labs nor any of its subsidiaries or affiliates shall be
>>> liable for the message if altered, changed or falsified.
>>> Ce message et toutes les pi?ces jointes (ci-apr?s le "message") sont
>>> confidentiels et ?tablis ? l'intention exclusive de ses destinataires.
>>> Toute utilisation ou diffusion non autoris? est interdite. Tout
>>> message ?lectronique est susceptible d'alt?ration. DxO Labs et ses
>>> filiales d?clinent toute responsabilit? au titre de ce message s'il a
>>> ?t? alt?r?, modifi? ou falsifi?.
>>>
>>>
>>> _______________________________________________
>>> Cbc mailing list
>>> Cbc at list.coin-or.org
>>> http://list.coin-or.org/mailman/listinfo/cbc
>> -------------- next part --------------
>> An HTML attachment was scrubbed...
>> URL:<http://list.coin-or.org/pipermail/cbc/attachments/20150206/0f597d30/attachment-0001.html>
>>
>> ------------------------------
>>
>> _______________________________________________
>> Cbc mailing list
>> Cbc at list.coin-or.org
>> http://list.coin-or.org/mailman/listinfo/cbc
>>
>>
>> End of Cbc Digest, Vol 92, Issue 5
>> **********************************
>
>
>
> _______________________________________________
> Cbc mailing list
> Cbc at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/cbc

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/cbc/attachments/20150208/16806228/attachment.html>