fastSTRUCTURE: logistic prior runs continuing infinitely

[Adam Cardilini]

G'day Guys,

I have had success running fastSTRUCTURE using simple priors and logistic priors. However, over the last few days I've been getting a problem where I run fastSTRUCTURE with a logistic prior and it never completes the analysis. It is particularly weird because I am using a script to do submit runs with different K's and runs with some K's work while others will not, e.g. a run with K6 gets caught up and never completes, while K8 runs and works.

Has anyone else had this issue? Does anyone know what might be going on?

Thanks for your help.

Cheers,

Adam 

[Vikram Chhatre]

Hi Adam,

Try lowering convergence criteria:

--tol=10e-5

I had a similar problem with logistic prior and this was the suggested solution. 

V

--
You received this message because you are subscribed to the Google Groups "structure-software" group.
To unsubscribe from this group and stop receiving emails from it, send an email to structure-softw...@googlegroups.com.
To post to this group, send email to structure...@googlegroups.com.
Visit this group at http://groups.google.com/group/structure-software.
For more options, visit https://groups.google.com/d/optout.

[Adam Cardilini]

Thanks for the tip Vikram, and thanks for always responding to my basic questions!

Unfortunately, reducing --tol= didn't help me. I even tried ridiculously low numbers (eg. 10e-1) just to see if it would finish. None of them stop processing. 

A simple prior suggests I have K = 2, but when trying to run this with a logistic prior it gets stuck. Does this mean I really don't have K = 2 or can I trust the output from the simple prior?

Thanks again,

Adam 

On Friday, August 8, 2014 9:05:57 AM UTC+10, Vikram Chhatre wrote:

Hi Adam,

Try lowering convergence criteria:

--tol=10e-5

I had a similar problem with logistic prior and this was the suggested solution. 

V

On Thursday, August 7, 2014, Adam Cardilini <a.car...@gmail.com> wrote:

G'day Guys,

I have had success running fastSTRUCTURE using simple priors and logistic priors. However, over the last few days I've been getting a problem where I run fastSTRUCTURE with a logistic prior and it never completes the analysis. It is particularly weird because I am using a script to do submit runs with different K's and runs with some K's work while others will not, e.g. a run with K6 gets caught up and never completes, while K8 runs and works.

Has anyone else had this issue? Does anyone know what might be going on?

Thanks for your help.

Cheers,

Adam 

--
You received this message because you are subscribed to the Google Groups "structure-software" group.

To unsubscribe from this group and stop receiving emails from it, send an email to structure-software+unsub...@googlegroups.com.
To post to this group, send email to structure-software@googlegroups.com.

[Vikram Chhatre]

Hi Adam,

Can you share your results with simple prior?  e.g. bar plot and the output of chooseK.py script?  Logistic prior is useful when structure in the data is weak. I have a feeling that fastSTRUCTURE is encountering some technical problem when running the logistic prior.  It should not be taking this long regardless of how difficult the solution might be.

V

To unsubscribe from this group and stop receiving emails from it, send an email to structure-softw...@googlegroups.com.
To post to this group, send email to structure...@googlegroups.com.

[Adam Cardilini]

G'day Vikram,

Please see below for the output of the chooseK.py script and the distruct.py plot. This is 'second level' data, that is, I found three genetic groups with my full data set. I then wanted to see if I could find substructure by running structure.py on each genetic group. This is one of those genetic groups.

I find it very odd that the plot below shows two groups that are so clearly distinct and it makes me suspect something weird is going on. Perhaps my data is getting messed up when I subset the original data set?

Any ideas would be really helpful.

Thanks again,

Adam

chooseK.py output:

Model complexity that maximizes marginal likelihood = 2

Model components used to explain structure in data = 1

V

To unsubscribe from this group and stop receiving emails from it, send an email to structure-software+unsubscribe@googlegroups.com.

To post to this group, send email to structure-software@googlegroups.com.
Visit this group at http://groups.google.com/group/structure-software.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "structure-software" group.
To unsubscribe from this group and stop receiving emails from it, send an email to structure-software+unsub...@googlegroups.com.

[Craig Anderson]

Did you manage to resolve this?

Thanks,

Craig

[Adam Cardilini]

G'day,

I believe it might be something to do with having SNPs with missing data for all individuals within a subgroup of the data. No work around that I can think of other than to remove the offending SNPs

Cheers,
Adam

[Madhav Bhatta]

Hi Adam,

Did you able the fix the issue of logistic prior? If yes, could you please share your experience on fixing this issue? I am having the same problem. I ran simple prior and it gave me result with in a minute but with logistic prior, I have no hope. It's already been 4 days running k=2 with nothing.
Any suggestions provided will be appreciated!!

Thank you,
Madhav

[Vikram Chhatre]

Long computation time does not mean failed run.  Note that logistic priors always take much longer to run than simple.  This route is primarily taken to detect weak population structure.  

--
You received this message because you are subscribed to the Google Groups "structure-software" group.
To unsubscribe from this group and stop receiving emails from it, send an email to structure-software+unsub...@googlegroups.com.

To post to this group, send email to structure-software@googlegroups.com.

Visit this group at https://groups.google.com/group/structure-software.

[Madhav Bhatta]

Thank you Vikram for you prompt response. How long will it take to run? I have 37,500 SNPs from 141 genotypes. Is there a way to speed up this process? I have been running for 4 days for just one run. 

Thank you again, 

Madhav.

You received this message because you are subscribed to a topic in the Google Groups "structure-software" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/structure-software/75XyLbD6hPU/unsubscribe.
To unsubscribe from this group and all its topics, send an email to structure-softw...@googlegroups.com.
To post to this group, send email to structure...@googlegroups.com.

[Madhav Bhatta]

Thank you Vikram for you prompt response. How long will it take to run? I have 37,500 SNPs from 141 genotypes. Is there a way to speed up this process? I have been running for 4 days for just one run. 

Thank you again. Madhav.

On May 16, 2017, at 3:19 PM, Vikram Chhatre <cryptic...@gmail.com> wrote:

Long computation time does not mean failed run.  Note that logistic priors always take much longer to run than simple.  This route is primarily taken to detect weak population structure.  

On Tue, May 16, 2017 at 2:16 PM, Madhav Bhatta <bhatta...@gmail.com> wrote:

On Thursday, August 7, 2014 at 4:59:54 PM UTC-5, Adam Cardilini wrote:
> G'day Guys,
>
>
> I have had success running fastSTRUCTURE using simple priors and logistic priors. However, over the last few days I've been getting a problem where I run fastSTRUCTURE with a logistic prior and it never completes the analysis. It is particularly weird because I am using a script to do submit runs with different K's and runs with some K's work while others will not, e.g. a run with K6 gets caught up and never completes, while K8 runs and works.
>
>
> Has anyone else had this issue? Does anyone know what might be going on?
>
>
> Thanks for your help.
>
>
> Cheers,
> Adam 

Hi Adam,

Did you able the fix the issue of logistic prior? If yes, could you please share your experience on fixing this issue? I am having the same problem. I ran simple prior and it gave me result with in a minute but with logistic prior, I have no hope. It's already been 4 days running k=2 with nothing.
Any suggestions provided will be appreciated!!

Thank you,
Madhav

--
You received this message because you are subscribed to the Google Groups "structure-software" group.

To unsubscribe from this group and stop receiving emails from it, send an email to structure-software+unsubscri...@googlegroups.com.

To post to this group, send email to structure-software@googlegroups.com.

Visit this group at https://groups.google.com/group/structure-software.

For more options, visit https://groups.google.com/d/optout.

-- 
You received this message because you are subscribed to a topic in the Google Groups "structure-software" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/structure-software/75XyLbD6hPU/unsubscribe.
To unsubscribe from this group and all its topics, send an email to structure-software+unsub...@googlegroups.com.