What Liviu means by this is that in your version of the evaluator,
all slaves' random number generators are hard-coded seeded with a
seed based on the wall clock time. ECJ will make sure the seeds are
different from one another. There is no option to do anything else
in the new master-slave evaluator.
Sean
On Jan 29, 2006, at 2:43 AM, I Jonyer wrote:
> I see, thanks.
> Where do you get N and where do you set it? Can I set it to 'time'?
>
> Istvan
>
>
>> From: Liviu Panait <[log in to unmask]>
>> Reply-To: ECJ Evolutionary Computation Toolkit <ECJ-
>> [log in to unmask]>
>> To: [log in to unmask]
>> Subject: Re: ECJ14 beta
>> Date: Sun, 29 Jan 2006 02:22:28 -0500
>>
>> Istvan,
>>
>> evalthreads does not have any impact on the new master-slave
>> evaluator. We support it in the new master-slave just for
>> compatibility with the other evaluators. If I were to use the
>> system, I'd set evalthreads=1
>>
>> As evaluators may come online and crash at any point in time,
>> their random seeds are usually set as N+K, where N is a fixed
>> number, and K is the number of slaves that have connected up to
>> now to the master.
>>
>> Best regards,
>>
>> Liviu.
>>
>> On Jan 28, 2006, at 11:17 PM, I Jonyer wrote:
>>
>>> Liviu (or Sean),
>>>
>>> I have been testing the distributed functionality in ECJ14 beta.
>>> I have been playing with the number of evaluation threads and
>>> the actual number of clients I am starting up. The number of
>>> evaluation threads does not seem to have any effect. The master
>>> always accepts all the clients that want to register, never
>>> refusing any of them or waiting for more. I like this the way it
>>> is, but I am wondering what, if any, role evalthreads plays in
>>> this version.
>>> Also, if I have 10 evalthreads specified, but 20 clients
>>> registering, what is the default value of the random seed for
>>> those clients which were never specified?
>>>
>>> Thank you,
>>> Istvan
|