Improved Accuracy and Better Input Data Size Estimations

• Improved accuracy for the flattened multiresolution / hierarchy levels
• Better estimation of the input nodes and clusters sizes when not specified in the header
• Standard shared CNL header parser
• Membership optional parameter added for more accurate estimation

The executable is built on Ubuntu x64 16.04

Much improved performance due to the algorithmic updates

• Much improved performance on large networks (orders reduced computational complexity) due to the redesigned stochastic sampling, should be really close to linear now
• New option (-a) for even faster (though less accurate) evaluations on large networks
• Refined notifications

The executable for built on Ubuntu 16.04 x64 is attached

Optimized to linear complexity O(nodes_num) rateining the accurracy

• Optimized to linear complexity O(nodes_num)

Binary for Ubuntu 16.04.1 x64 is attached

Fast and Accurate reimplementation

• Execution became about 2 ORDERS faster on small network (but the complexity is O(cls_num^2))
• Memory consumption reduced twice
• Accuracy improved on large networks by processing the whole table of relations instead of the predefined 8K community pairs

The code significantly cleaned up.

Node base synchronization and FNMI added, parsing fixed

• Optional synchronization of the node base for the fair evaluation of top K ground-truth communities
• Optional "Fair" NMI added (anyway standard NMI is more meaningful and accurate)
• Accuracy fixes in the NMI evaluation
• Parsing fixed for CNL, original gecmi format dropped

Initial Release

• Fuly automated bulid using single Makefile is made
• Removed Python wrapper and related dependancies => single executable is produced
• Chunking in TBB is tunned => ~ 1.5 times performance improvement
• Minor fixes