Stat Test for other models

[vivekaxl]

GD-New.pdf IGD-New.pdf Evals-New.pdf

[vivekaxl]

SaC (58 Dimensions):

• All the method have same ranks wrt. to GD and IGD while AL2 uses the least number of evaluations.

  Preferred Method: AL2

[vivekaxl]

x264 (17 Dimensions)

• MMRE cannot be our preferred method since the number of evaluations is always greater than 11k points which is 50 times more that ALX approaches.
• AL1 outperforms AL2 in all cases except for x264-5_6. But it uses twice as much evaluations wrt. to AL2.

Preferred Method: AL1

[vivekaxl]

POMX (9 Dimensions)

• All the techniques with exception of MMRE perform similarly in all cases except for POM3C, where AL1 outperform Rank and AL2.

Preferred Method: AL2

Since the number of evaluations required by AL2 is always half of the number of evaluations used by AL1. In case of POM3C, number of evaluations used by AL2 is 8 times less than AL1.

[vivekaxl]

XOMOX (27 Dimensions)

• AL1 and AL2 consistently performs the best wrt. to GD and IGD.
• Number of evaluations used by AL2 is always less than AL1. The win factor of AL2 wrt. to AL1 ranges from 1.22 to 6.96.

Preferred Method: AL2

[vivekaxl]

MONRP (50 Dimensions)

• AL1 cannot be our preferred method since it uses the most number of evaluations 200 times more evaluations than AL2.
• Rest of the methods perform similarly (same rank) but AL2 uses much less evaluations when compared to other methods.

Preferred Method: AL2.

[vivekaxl]

TODO:

More conclusive and concrete result would be compare the performance of AL2 with other optimizers such as NSGA2, SPEA2 and SWAY. ETA: Thursday Evening. #20