# Re: [AMBER] ABMD with multiple walkers

From: Qinghua Liao <scorpio.liao.gmail.com>
Date: Tue, 29 May 2018 09:31:31 +0200

Hello Feng,

Thanks a lot for your clarification. I understand it now!

All the best,
Qinghua

On 05/29/2018 03:01 AM, Feng Pan wrote:
> Leading to 0 walker means this replica will be removed, replaced by another
> replica which leads to 2 walkers.
>
> Leading to 1 walker means the walker will stay with itself, but the index
> may
> shift to the next one when it has neighboring replica leading to 0 walkers.
>
> When it is converged, each replica should lead to 1 walker, which means
> the selection actually does not take effects.
>
> Feng
>
>
> On Mon, May 28, 2018 at 8:36 AM, Qinghua Liao <scorpio.liao.gmail.com>
> wrote:
>
>> Hello Feng,
>>
>>
>> A few more questions:
>>
>> Leading to 0 walker means that the walker will be stayed with itself or
>> copied by another walker
>> which leads to 2 walkers?
>> Leading to 1 walker means that this walker might be stayed with itself or
>> can copied by another walker which leads to 1 walker too?
>>
>> What would be the selection scores of all walkers if the ABMD with
>> multiple walkers is converged?
>>
>> Thanks so much!
>>
>>
>> All the best,
>> Qinghua
>>
>>
>>
>> On 05/28/2018 04:05 AM, Feng Pan wrote:
>>> Hi, Qinghua
>>>
>>> Yes, you are right, I made a mistake. I check the codes. A random number
>>> between 0 to 1/8
>>> should be added to the score.
>>> For example, for the first replica, if the sum is below 1/8 (like the
>> score
>>> is below 1/8 and
>>> the random number is also small), this replica leads to zero walkers. If
>>> the sum is between
>>> 1/8 and 2/8, leads to 1 walker. The following replicas will be calculated
>>> to make sure the
>>> total number of walkers is 8.
>>> So basically when you see the score is below 1/8, the walker could be 0
>> or
>>> 1.
>>>
>>>
>>> When I say walkers will be copied, I mean the coordinates
>>> will be copied.
>>>
>>> Best
>>> Feng
>>>
>>> On Sun, May 27, 2018 at 5:08 PM, Qinghua Liao <scorpio.liao.gmail.com>
>>> wrote:
>>>
>>>> Dear Feng,
>>>>
>>>> Thanks a lot for your explanation.
>>>>
>>>> Following your explanation, I also found that there are some walkers
>>>> which have the scores
>>>> between 1/8 and 2/8, but the walkers were not copied to 2 walkers.
>>>>
>>>> When the score is less than 1/8, the walker will be kept itself, then
>>>> the sign is "=> 0" or "=> 1"
>>>>
>>>> About the term "walkers will be copied", are coordinates copied or
>>>> selection score (weight)? Thanks a lot!
>>>>
>>>>
>>>> All the best,
>>>> Qinghua
>>>>
>>>> On 05/27/2018 04:16 AM, Feng Pan wrote:
>>>>> Hi, Qinghua
>>>>>
>>>>> the selection score is the ratio of the replica selection weight over
>> the
>>>>> total weight,
>>>>> so if you have 8 replicas here, and the score is below 1/8, the walker
>>>> will
>>>>> keep
>>>>> itself. If the score is between 1/8 and 2/8, the walker will be copied
>>>> to 2
>>>>> walkers, as
>>>>> shown like => 2 walkers.
>>>>>
>>>>> When the entropy is above the threshold, the selection will be
>>>> calculated,
>>>>> but if
>>>>> every selection score is below 1/8, there will be no resampling, every
>>>>> walker will
>>>>> be kept just itself.
>>>>>
>>>>> Best
>>>>> Feng
>>>>>
>>>>> On Fri, May 25, 2018 at 7:00 PM, Qinghua Liao <scorpio.liao.gmail.com>
>>>>> wrote:
>>>>>
>>>>>> Hello Amber developers,
>>>>>>
>>>>>> Currently, I am running ABMD with multiple walkers, but I have some
>>>>>> questions regarding interpreting the output.
>>>>>> Here is my input for the multiple walkers:
>>>>>>
>>>>>> selection_freq = 500
>>>>>> selection_constant = 0.00001
>>>>>> selection_epsilon = 0.0
>>>>>>
>>>>>> And here are some output from the simulations:
>>>>>> #
>>>>>> NFE : selection score for walker 1 is 0.930991 / 7.992304 = 0.116
>> =>
>>>> 0
>>>>>> walker(s)
>>>>>> NFE : selection score for walker 2 is 0.881570 / 7.992304 = 0.110
>> =>
>>>> 1
>>>>>> walker(s)
>>>>>> NFE : selection score for walker 3 is 0.945704 / 7.992304 = 0.118
>> =>
>>>> 1
>>>>>> walker(s)
>>>>>> NFE : selection score for walker 4 is 0.951837 / 7.992304 = 0.119
>> =>
>>>> 1
>>>>>> walker(s)
>>>>>> NFE : selection score for walker 5 is 0.973335 / 7.992304 = 0.122
>> =>
>>>> 1
>>>>>> walker(s)
>>>>>> NFE : selection score for walker 6 is 1.103148 / 7.992304 = 0.138
>> =>
>>>> 1
>>>>>> walker(s)
>>>>>> NFE : selection score for walker 7 is 0.967854 / 7.992304 = 0.121
>> =>
>>>> 1
>>>>>> walker(s)
>>>>>> NFE : selection score for walker 8 is 1.237866 / 7.992304 = 0.155
>> =>
>>>> 2
>>>>>> walker(s)
>>>>>> NFE : Selection entropy 0.005597 is greater than threshold
>> 0.000000
>>>>>> NFE : Selection resampling : new 1 comes from 2
>>>>>> #
>>>>>> #
>>>>>> NFE : selection score for walker 1 is 1.042757 / 7.817606 = 0.133
>> =>
>>>> 1
>>>>>> walker(s)
>>>>>> NFE : selection score for walker 2 is 0.956335 / 7.817606 = 0.122
>> =>
>>>> 1
>>>>>> walker(s)
>>>>>> NFE : selection score for walker 3 is 0.775833 / 7.817606 = 0.099
>> =>
>>>> 1
>>>>>> walker(s)
>>>>>> NFE : selection score for walker 4 is 0.906343 / 7.817606 = 0.116
>> =>
>>>> 1
>>>>>> walker(s)
>>>>>> NFE : selection score for walker 5 is 0.938005 / 7.817606 = 0.120
>> =>
>>>> 1
>>>>>> walker(s)
>>>>>> NFE : selection score for walker 6 is 0.851424 / 7.817606 = 0.109
>> =>
>>>> 1
>>>>>> walker(s)
>>>>>> NFE : selection score for walker 7 is 1.232238 / 7.817606 = 0.158
>> =>
>>>> 1
>>>>>> walker(s)
>>>>>> NFE : selection score for walker 8 is 1.114672 / 7.817606 = 0.143
>> =>
>>>> 1
>>>>>> walker(s)
>>>>>> NFE : Selection entropy 0.009795 is greater than threshold
>> 0.000000
>>>>>> #
>>>>>>
>>>>>> How should I understand the selection score, how is it calculated?
>> What
>>>>>> does "=>0 (1,2) walker(s)" mean?
>>>>>> For the first assessment, the selection entropy is 0.005597,
>>>>>> which is greater than the threshold 0.0, and then there is a selection
>>>>>> resampling (new 1 comes from 2). For the second
>>>>>> assessment, the selection entropy is also greater than the threshold,
>>>>>> but there is no resampling. Why is this different?
>>>>>>
>>>>>> I appreciate any of your responds.
>>>>>>
>>>>>>
>>>>>> All the best,
>>>>>> Qinghua
>>>>>>
>>>>>> _______________________________________________
>>>>>> AMBER mailing list
>>>>>> AMBER.ambermd.org
>>>>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>>>>
>>>> _______________________________________________
>>>> AMBER mailing list
>>>> AMBER.ambermd.org
>>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>>
>>>
>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>
>

_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue May 29 2018 - 01:00:02 PDT
Custom Search