. Author manuscript; available in PMC: 2020 Apr 7.

Published in final edited form as: Adv Neural Inf Process Syst. 2019 Dec;32:10552–10563.

Table 1:

Graph statistics of generated formulas (mean ± std. (relative error to training formulas)).

Method	VIG		VCG			LCG
Method	Clustering	Modularity	Variable α_v	Clause α_c	Modularity	Modularity
Training	0.50±0.07	0.58±0.09	3.57±1.08	4.53±1.09	0.74±0.06	0.63±0.05

CA	0.33±0.08(34%)	0.48±0.10(17%)	6.30±1.53(76%)	N/A	0.65±0.08(12%)	0.53±0.05(16%)
PS(T=0)	0.82±0.04(64%)	0.72±0.13(24%)	3.25±0.89(9%)	4.70±1.59(4%)	0.86±0.05(16%)	0.64±0.05(2%)
PS(T=1.5)	0.30±0.10(40%)	0.14±0.03(76%)	4.19±1.10(17%)	6.86±1.65(51%)	0.40±0.05(46%)	0.41±0.05(35%)
G2SAT	0.41±0.09(18%)	0.54±0.11(7%)	3.57±1.08(0%)	4.79±2.80(6%)	0.68±0.07(8%)	0.67±0.03(6%)