ENH-relational-allocmodel

View source
ENH-relational-allocmodel
  • Contributors
    1. Loading...
Author Commit Message Date Builds
44 commits behind master.
Mike Hughes
ENH Updated allocation models for relational. Use standard notation of NodeStateCount, and (for assortative) we use entropy vector that is size K called Hresp_fg, plus a scalar Hresp_bg
Mike Hughes
FIX Monks dataset now working (single relation only), and small fix to FiniteAssortativeMMSB so init is not broken. Still need some careful attention to initialization.
Mike Hughes
MAINT small cleanup and improvements to GraphXData and Monks.py
Mike Hughes
MAINT updating BernObsModel to pep8 compliance, improved readability
Will Stephenson
FIX kmeans and randexamplesbydist in FromScratchRelational now work (Mike, is this what you intended randexamplesbydist to do?)
Will Stephenson
ADD Lawyers dataset. Also needs its .mat file edited to work with new GraphXData
Will Stephenson
ADD Monks dataset. Need to edit rawData/Monks.mat to get working with new GraphXData
Will Stephenson
FIX Proper error GraphXData, ToyMMSBK6Gauss passes in data correctly
Mike Hughes
INPROGRESS adding initLP to init_global_params of each MMSB model
Mike Hughes
ENH HDPAssortativeMMSB added
Mike Hughes
ENH added HDPMMSB, which seems to work OK (ELBO doesnt drop in a very preliminary test)
Mike Hughes
MAINT not supporting undirected MMSB in near future, so removing for now...
Mike Hughes
ENH FiniteAssortativeMMSB now does correct entropy calculation, as written in Will's thesis. Also, RelationalViz has some fixes to plot both Assortative and nonAssortative results
Mike Hughes
FIX FiniteMMSB init_global_params now breaks symmetry by adding small random values in initNodeStateCount to self.alpha
Mike Hughes
FIX ToyMMSBK6 dataset can cast nNodes from str to int
Mike Hughes
ENH started alternative computation of entropy for assortative models. TODO: make it correct, so Hresp equals Hresp2 in line 175
Mike Hughes
ENH added FiniteAssortativeMMSB, which is mostly functional (need to fix entropy calc to include bg terms
Mike Hughes
ENH GraphXData is now defined by attribute nEdgesTotal, which indicates whole-dataset size.
Mike Hughes
ENH FiniteMMSB works with Bern likelihood. Demo: bnpy.Run ToyMMSBK6 FiniteMMSB Bern VB --nLap 3
Mike Hughes
MAINT name change from FiniteaMMSB to FiniteAssortativeMMSB, so its easier to tell them apart and no one gets confused.
Mike Hughes
ENH GraphXData refactored totally, so represents dense graphs only. Will make SparseGraphData object later.
Mike Hughes
MAINT cleanup formatting for HDPaMMSB and Toy dataset files
Mike Hughes
MAINT ToySMSBK6 and ToyMMSBK6 have nice adj matrix visualizations now. Networkx viz disabled for these two scripts, since it doesnt seem to show much useful.
Will Stephenson
FIX some functionality in RelationalViz now working with new GraphXData
Will Stephenson
Edited GraphXData docstring, fixed ToyMMSBK6 and ToyMMSBK6Gauss
Mike Hughes
MAINT now HDPaMMSB and MMSBUtil files are pep8 compliant. hopefully didn't ruin meaning of any code.
Mike Hughes
MAINT Finite versions of all relational models respect PEP8 formatting
Mike Hughes
ENH reformating all allocmodel/relational/ files to be PEP8 compliant
Mike Hughes
ENH cleaned up ToyMMSBK6 pretty decently. Still need to fix relational viz to make that visualization possible with new edgeSet only GraphXData
Mike Hughes
MAINT integrated SMSB code. all comply with PEP8 now.