Efficient discovery of overlapping communities in massive networks PK Gopalan, DM Blei Proceedings of the National Academy of Sciences 110 (36), 14534-14539
I got puzzled when reading this code in iterations of the function "PhiComp::update_phis(bool is_phi1)" in mmsbinfer.hh. According to the paper, when y==0,u is supposed be (1 - b[k]) *log(1-epsilon), but this case is not considered here.
I would like to ask if the node IDs used in the communities.txt output file of the algorithm are the external IDs the nodes had in the input file provided by the user, or the IDs that the algorithm internally assigns to each node.
Compiling using Mac OS 10.9 gives the following warning:
"In file included from main.cc:6:
./fastamm2.hh:222:38: warning: implicit conversion from 'double' to
'unsigned int' changes value from 0.99 to 0 [-Wliteral-conversion]
static const unsigned int THRESH = 0.99;"
Dear Prem,
When I plot the third column of Heldout.txt (which according to the detailed_readme, it should be the average held-out log likelihood), I see there are like overshoots before reaching to the convergence. I was wondering if it shouldn't be monotonically increasing (or at least after few first iterations)?
Attached is an example. I set the number of iterations to be 500.
Thanks,
I'd like to improve the install instructions with specific package names for dependencies. I'm not in a pposition to look into it properly right now, but here's my experience for the record:
On Ubuntu (12.04) i had to install
libgsl0-dev
libgsl0ldbl (libraries) was already installed, not sure if it's in the base distro?
gsl-bin - I installed it, but dont' think it's needed..
probably one of the blas packages is needed to - openblas/atlas
The class member variable "link_thresh" in class Env is never assigned and always set to default 0.
In the constructor of class Env , the passed parameter has the same name "link_thresh". Although this parameter "link_thresh" is set to 0.5, it is not assigned to the class member variable "link_thresh", since the later is often used in other codes.
The same cases to other class member variable "lt_min_deg" and "gap".
Assume that we generated θ and B matrix, where θ[i] is a vector of community memberships and B is the community-community interaction matrix. In another words, P = Theta * B *Theta.T, and A = Bernoulli(P).
I saved A to tab-separated file "network.txt".
How should i run your script to get recovered Theta and B matrices?