Can I calculate likelihood ratios if I calculated get_edges_prob in different runs?

classic Classic list List threaded Threaded
3 messages Options
P-M
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Can I calculate likelihood ratios if I calculated get_edges_prob in different runs?

P-M
I am trying to run "get_edges_probs" for a large number of edges on a reasonably sized network. This would take too long to run on my local machine so I am considering sending it off to a cluster. Speaking to the admin I would need to split the job into several smaller jobs, each of which I can then run on a different node.

My question is thus: If I split my list of edges into, say, 50 separate lists and then set up 50 jobs, each of which processes one of these lists, am I able to compare the results of these jobs to each other by simply recombining the resulting set of loglikelihoods to then calculate my likelihood ratios as outlined in the cookbook? Or does the stochastic nature of the algorithm mean that I can't necessarily compare results from different runs to each other in this manner?

Thank you for any help in advance.

Best wishes,

Philipp
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Can I calculate likelihood ratios if I calculated get_edges_prob in different runs?

Tiago Peixoto
Administrator
On 16.06.2017 07:35, P-M wrote:

> I am trying to run "get_edges_probs" for a large number of edges on a
> reasonably sized network. This would take too long to run on my local
> machine so I am considering sending it off to a cluster. Speaking to the
> admin I would need to split the job into several smaller jobs, each of which
> I can then run on a different node.
>
> My question is thus: If I split my list of edges into, say, 50 separate
> lists and then set up 50 jobs, each of which processes one of these lists,
> am I able to compare the results of these jobs to each other by simply
> recombining the resulting set of loglikelihoods to then calculate my
> likelihood ratios as outlined in the cookbook? Or does the stochastic nature
> of the algorithm mean that I can't necessarily compare results from
> different runs to each other in this manner?
The edge probability does not factorize over the edges, so the product of
the probability of the different lists will not be the same as the
probability for the whole list.

If the set of edges is much smaller than the total number, this could be
true _approximately_, but it can introduce errors in your analysis.

Best,
Tiago

--
Tiago de Paula Peixoto <[hidden email]>


_______________________________________________
graph-tool mailing list
[hidden email]
https://lists.skewed.de/mailman/listinfo/graph-tool

signature.asc (849 bytes) Download Attachment
--
Tiago de Paula Peixoto <tiago@skewed.de>
P-M
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Can I calculate likelihood ratios if I calculated get_edges_prob in different runs?

P-M
OK, thank you Tiago, that is useful to know!

Best,

Philipp
Loading...