MCL_lite is a memory-efficient implementation of Markov Clustering Algorithm for large-scale networks. In our tests, MCL_lite can cluster a network with 2 billion edges on a desktop with 16GB RAM while the original MCL needs more than 100GB RAM.
Currently, MCL_lite only support 64-bit linux operation system. Make sure that you have the following installed:
- Python2.7 (Recommend Anaconda)
-
Install dependency packages via pip:
$ pip install scipy numpy scikit-learn
-
Install dependency packages via conda:
$ conda install scipy numpy scikit-learn
$ git clone https://github.com/Rinoahu/MCL_lite
$python MCL_lite/mcl_sparse.py -i foo.xyz -I 1.5 -o foo.mcl -a 8 -m 8 -d t
-i: input network in 3 column tab-delimited format. The fist and second columns stand for nodes and third column stands for weight between these two nodes. For example:
N0 N1 1.5
N0 N2 1.2
N0 N3 1.1
...
-I: float. inflation parameter for mcl
-a: int. cpu number
-o: str. name of output file
-d: t|f. is the graph directed? Default is t(true)
-g: int. how many gpus to use for speedup. Default is 0
-r: t|f. resume the work after crash. Default is f(false)
-m: int. memory usage limitation. Default is 4GB