aburnap / mixed-membership-stochastic-blockmodel Goto Github PK

Hi,

first thanks for making this PyMC implementation of the MMSB available! It is very helpful to get started with implementing latent space models in PyMC.

I have downloaded the code and tried to run the example in MMSB_run.py
I have encountered the following problems.

I am using python version 2.7.3-0ubuntu7.1
and numpy version 1:1.6.2-1ubuntu1 .

1. Data file is not loaded correctly.

The data file in the example does not get loaded with the existing code. The problem is that numpy splits values on whitespace by default and the text file uses comma (,).
The problem is easily fixed by specifying the delimiter explicitly:

data_matrix=np.loadtxt("../data/Y_alpha0.1_K5_N20.txt", delimiter = ",")

2. MCMC model is empty

When running the mcmc sampler for the MMSB, the set of variables in the MCMC model seems to be empty!! The sampler runs through without error messages (even very fast) but does not seem to do anything. As a result there is no output statistics after the sampling.

Here is the step-by-step code that I execute to create the sampler class and run it.

From MMSB_run.py :

$ raw_model = model.create_model(data_matrix, num_people, num_groups, alpha, B)

$ MC_model = pymc.Model(raw_model)

$ print MC_model.nodes // There are variables here
set([<pymc.distributions.Multinomial 'z_3T4_vector' at 0x6163a10>, <pymc.distributions.Multinomial 'z_12T17_vector' at 0x6978050>, <pymc.distributions.Multinomial 'z_16T3_vector' at 0x6c9f4d0>, <pymc.distributions.Multinomial 'z_12T3_vector' at......

$ M = pymc.MCMC(model)

$ print M.nodes // Suddenly there are NO variables in M ??!!!
set([])

$ M.sample(10000,500, thin=5)
[_100%_**] 10000 of 10000 complete

$ pymc.Matplot.plot(M)
// no output here

$ M.summary()
// no output here either

I have not been able to fix the problem yet. Any help would be appreciated.

Thanks,

Daniel

.....