#RGRAPHM

This is a C++ version of the rpgrah program. The original can be found at seeslab.net or github.com/seeslab/rgraph

Paper:

Predicting Human Preferences Using the Block Structure of Complex Social Networks

In this version, the main objective is performance. If you find any bug or something to improve, please contact me at manu.mirandad@gmail.com or open an issue in the rgraphm repository.

###USAGE

main_recommender -t trainFile -q queryFile -s randomseed -m mark -a [gibbs]

-a: Algorithm to perform the calculations. Options are "metropolis"
    or "gibbs". Default is "metropolis".

-t: Path to train file in format a b c where 'a' is the first 
    node id, 'b' is the second node id and c is the weight of their 
    link. Weight values must range from 0 to N.

-q: Path to query file in format a b where 'a' is the first 
    node id of the query and 'b' is the second node id of the query.

-s: Seed for the random. If you use the same value and number 
    of iterations, you will always obtain the same results.

-m: Number of different ratings a user can apply. For example, if a 
    node of the second set can be voted from 0 to 4 stars, mark would 
    be 5.

-n: Numer of Monte Carlo steps.

-h: Show help.

###DEPENDENCIES

• Execution

• libgsl0dbl
• boost

• Compilation

• libgsl0-dev
• boost-devel

###PERFORMANCE

To check performance and improve it, you can use valgrind profiler (apt-get install valgrind). Once installed, execute the application as follows:

valgrind --tool=callgrind ./main_recommender args

After execution ends, an output file is generated. You can open it directly and try to interpret it. I recommend using Kcachegrind (apt-get install kcachegrind) which provides an awesome interface to navigate through call graphs, source code (if available) and more.

NOTE: If you want to generate profiling info will valgrind, the program will last much more. You also need to compile it with debug options (-g in gcc).

• Don't try to access gglinks.data like a 1D array, it is slower.
• Removing the set_ind switch in calculatedH doesn't improve performance.