As far as I’m aware, an R instance on EC2 is just like an R instance on your computer. In other words, running a single instance of R without any form of parallell processing package uses a single core (at least for now).

Although there are parallell processing packages, I have found that for one-off analysis is that the potential gain is limited (e.g., the cost of extra programming time is often greater than the cost of extra running time). If you are to set up a repeating computationally intensive task, you might want to consider reprogramming it in c++ or using a map-reduce job (i.e., Hadoop / Amazon Elastic MapReduce).

Let me know what you end up doing!

Good luck!
Tore

When running R code over multiple threads on EC2, do you need to use snow, foreach, or another parallellization package? Or can I just run my normal code and see the speed go up as the number of cores increases in the cloud? In other words, do I need to learn additional programming tools in order to run R in the cloud?

thanks,
Peter

Glad you are finding it useful!

The link reshuffling is not very efficient as it relies on igraph. I used to have a much more efficient script for sparse graphs, but had to provide loads of support for people using it on dense graphs. If you send me an email, we can see if this might be more appropriate for you.

Best,
Tore

I wondered how well the algorithm scaled to larger networks? I have a network of over 10k nodes and about 80k edges, I have been running your program for a while so far, but wondered if you had any idea on how long it might take to calculate?

Many thanks, and again, great site.

Great that you managed to solve it! It is easier to search for the ami id. If you go to the bioconductor page, they list their latest version (currently ami-9f12dcf6; R-2.14).

Best,
Tore

I had searched for an instance using the “RStudio” keyword rather than bioconductor.
Once I got the BioConductor instance it all started working.

Thanks again.

- john

I am having some trouble though. Evertything is fine until I try to do the tunneling to RStudio.
I suspect there is a deeper problem since I even tried opening port 8787 like the Bioconductor doc says and still could not get my browser to open RStudio.

Is it possible that some instances have RStudio listening on another port, e.g. port 80?
I was trying to figure out whether the issue was on my PC (Windows XP but running ZoneAlarm
for a firewall and other security measures) or on the server side.
On the EC2 instance I ran netstat and the message seemed to suggest that RStudio
was listening on port 80. Still unable to connect on port 80 though with my web browser.

Anyone else having these issues?

thanks again,
– john

Thanks for letting me know. I just tried all of them, and they seemed to work. As they are all hosted on the same server, they should all work or all not work (e.g., if the server is down). Please email me if you have further issues.

Tore

Thank you for your comment. I tried to finish the two-mode section before the holidays. Just this page and the random networks pages left. Hopefully I will try to get this completed soon.

I am not aware of a specific local structural holes measure for two-mode networks. I have an idea for one based on a similar concept as the two-mode local clustering coefficient. At the moment, I am uncertain how well this measure will perform. As such, you might want to use the local clustering coefficient for two-mode networks (the local clustering coefficient is in many ways a simple constraint measure).

What do you think?
Tore

All best,
Vaclav

I am not entirely sure how Pajek does two-mode projection, but here is the documentation for the projectiom_tm-function in tnet: tnet » Two-mode Networks » Projection

Best,
Tore

I can see the mathematical point of the examples (A,B) and (B,C) with both line values set to 2 versus having one set to 3 and the other to 1. I’m still thinking about what the sociological or other implications might be.

Thanks for your comment!

Indeed it would be great. If I do get time to program it, I will. Do you have any code for weighted eigenvector centrality?

Best,
Tore

This definitely used to be the case with Anchorage as you said. The data used in the post is from 2010 (BTS) and August 2011 (Openflight), so it should have a limited impact. Although there might be some smaller planes which have to refuel, the aim of the post was not to analyse the airport network, but to highlight two statistical problems: sample selection and sensitivity to measurement using a dataset where “true”-data exist.

“True” data is immensely important when proposing new methods in efforts to validate them. For the airport network, it is possible to use data from the Airports Council International (2011). This data states that large cities are the ones with the highest throughput of international passengers. Admittedly, by using transit of international passengers, US airports fall on the ranking as the US has a very high level of domestic air travel.

An interesting extension of weighted betweenness is a measure that does not assume all nodes to be equal. For example, there are less people who live in Oslo than Los Angeles, and who subsequently fly. As such, the shortest path connecting major metropolitan areas might be more important than the one connecting two smaller cities. In turn, a node on the former path should be assigned a higher score than a node on the latter one. By weighting both ties and nodes, the new measure might be more appropriate than current ones. Any thoughts on this measure?

Best,
Tore

Thanks! The edgelist format does indeed not include isolates when they are at the end of the node id sequence. If you have a look at the example in the Closeness centrality in networks with disconnected components-blog post, you can see how you can include isolates by making sure a non-isolate has the highest node id.

Best,
Tore

Great that you got it to work!

Hope you’ll find it useful.

Tore

Look forward to hear from you
Thanks so much