FeatureModels

[vivekaxl]

Experiment:

• Sample 100 points with replacement
• Generate 50 scalarized objective scores (since ELA for multiobjective is not available)
• Calculate 30 features from the scalarized datasets
• Do N way cross eval. Learn from ~test-dataset and predict test-dataset.

Datasets	MMRE (Objective Intrinsic D)	IQR	#points
SQLite	1.19	0.33	418
x264	2.14	0.84	1165
BerkeleyDBC	5.91	0.76	2561
clasp	7.11	0.26	701
BerkeleyDBJ	9.45	0.65	181
WGet	10.48	1.72	189
Apache	10.8	2.1	193
EPL	12.35	0.42	366
LinkedList	13.36	1.89	205
BerkeleyDB	13.63	0.48	2561
lrzip	20.15	0.26	433
BerkeleyC	27.15	2.98	257
AJStats	37.72	0.67	30256

[vivekaxl]

Objectives

• Performance Measure
• Number of Features used