Skip to content

Releases: eXascaleInfolab/GenConvNMI

Optional Node Base Sync to the 1st File

02 Nov 19:52
Compare
Choose a tag to compare

Features

  • Optional node base synchronization to the first file added besides the synchronization to the minimal node base
  • Description refined

Fixes

The executable is built on Ubuntu x64 16.04

Accuracy and Gonvergence Improved

15 Sep 16:21
Compare
Choose a tag to compare

Accuracy and convergence on large datasets improved further insignificantly affecting the efficiency.

The executable is built on Ubuntu x64 16.04.

Heavy Overlaps Processing Refined and Optimized

12 Dec 12:20
Compare
Choose a tag to compare
  • Normalization of the vertex samples (importance) refined
  • Multimatch of the modules refined and optimized (hard and soft matching converged!)
  • Some speed optimizations

The executable is built on Ubuntu x64 16.04

NMI Constraints Considered

06 Dec 15:09
Compare
Choose a tag to compare
Pre-release
  • NMI constraints considered: explicit exceptions are throw when NMI is not applicable instead of outputting 0 or 1. For example, NMI is formally equal to zero when one clustering contains a single cluster of all nodes and another clustering can be any, which says nothing about the clusterings similarity
  • Some accuracy and performance optimizations made to avoid evaluation complexity dependence from the overlaps size
  • Code refined

The executable is built on Ubuntu x64 16.04

Evaluation of duplicated and equally similar clusters

04 Dec 21:12
Compare
Choose a tag to compare
  • Evaluation of equally similar clusters in the collections (including duplicated clusters, which earlier caused exceptions)
  • Bug fixes: correct evaluation of fully matched clusters when MI and H1/2 are zero now correctly yields 1, etc.
  • Minor optimizations

The executable is built on Ubuntu x64 16.04

Ids Remapping Added

04 Sep 02:56
Compare
Choose a tag to compare
  • Input Ids remapping added to allow non-solid range of arbitrary ids (still numbers), -i input option should be specified
  • Default input ids are allowed to start from 1 besides 0

By default (without -i option) the input ids should form the solid range that starts from 0 or 1.

The build is made on Linux Ubuntu 16.04 x64.

Known issues.

NMIsqrt Added

20 Jul 09:23
Compare
Choose a tag to compare
  • NMIsqrt added
  • Output options extended

The executable is built on Ubuntu x64 16.04

Node Estimation Fixed

23 Feb 12:31
Compare
Choose a tag to compare
  • Node estimation fixed to avoid "bad alloc" on memory over reserving
  • Minor refinements of notifications and docs
  • Common code with onmi is shared

The executable is built on Ubuntu x64 16.04

Improved Accuracy due to the Weighted Sampling

13 Feb 13:16
Compare
Choose a tag to compare
  • Improved accuracy due to the weighted sampling proportional to the number of clusters in the collection
  • Random generator parameterized with the random device is used instead of the random device

The executable is built on Ubuntu x64 16.04

Collections Oder Invariant Sampling

13 Feb 04:28
Compare
Choose a tag to compare

The stochastic sampling fixed to be collections order invariant

The executable is built on Ubuntu x64 16.04