Implementation for the Kaggle competition "Learning Social Circles in Networks".
Run python driver.py -h for help and options.
Flags:
- -s Compute statistics on data.
- --trim Output data common per egonet. This set will be ignored when displaying intersection attributes.
- -p Predict social circles using the specified predictor. Supported
predictors:
- igraph - Use various community detection algorithms from the igraph package. Can be combined with --edge parameter.
- kmeans - K-means clustering
- mcl - Markvoc clustering algorithm. Can be combined with --edge parameter.
- -v Visualize data. By default uses original topology to construct graphs.
- --split Split visualizations by circle.
- --save Save output. Graphical output is saved to the folder 'graphs' in the current directory.
- --show Show output during visualization calculations.
- --edge Select the edge function to use when visualizing data. Supported
options are:
- top: Uses the original graph topology.
- top-intersect: Uses original graph topology with a minor weight given to attributes that are in common.
- sim: Creates edges between similar users.
- tri: Creates edges between users with friends in common.
- combo: Uses both 'tri' and 'sim' to create edges.
- --prune Select a post-cluster-prediction pruning method. Supported options are:
- small: Remove small clusters.