Here is the feature subset we used in our CP06 paper "Performance Prediction and Automated Tuning of Randomized and
Parametric Algorithms".
It is actually only 43 features, not 46 as stated in the paper, sorry about that. We'll correct it in the final version of
the paper.
As a preprocessing step in the machine learning, constant features were removed (features for which the maximum and mimimum
value differed by no more than 10^(-10).)
Software to quickly extract these features from any given SAT instances will be available soon.
nvars
nclauses
vars_clauses_ratio
VCG_VAR_min
VCG_VAR_max
VCG_VAR_entropy
VCG_CLAUSE_mean
VCG_CLAUSE_coeff_variation
VCG_CLAUSE_max
VCG_CLAUSE_entropy
POSNEG_RATIO_CLAUSE_mean
POSNEG_RATIO_VAR_stdev
POSNEG_RATIO_VAR_max
BINARY_PLUS
TRINARY_PLUS
HORNY_VAR_coeff_variation
HORNY_VAR_min
HORNY_VAR_max
HORNY_VAR_entropy
horn_clauses_fraction
VG_mean
VG_coeff_variation
VG_min
vars_reduced_depth_4
vars_reduced_depth_16
vars_reduced_depth_64
saps_BestSolution_Mean
saps_BestSolution_CoeffVariance
saps_BestStep_Mean
saps_BestStep_CoeffVariance
saps_AvgImproveToBest_Mean
saps_AvgImproveToBest_CoeffVariance
saps_FirstLMRatio_Mean
saps_FirstLMRatio_CoeffVariance
saps_BestCV_Mean
gsat_BestSolution_Mean
gsat_BestSolution_CoeffVariance
gsat_BestStep_Mean
gsat_BestStep_CoeffVariance
gsat_AvgImproveToBest_CoeffVariance
gsat_BestCV_Mean
lobjois_mean_depth_over_vars
lobjois_log_num_nodes_over_vars