Table 5.

Pairwise comparison for positioning speed using “Tukey-Kramer” for a 95% confidence interval on a two-way ANOVA.

	Group 1	Group 2	Lower	Mean diff.	Upper	p-Value	Mean diff (%)
After learning	R	R + P	−0.5872	−0.2549	0.0774	0.1992	−3.04
	R	R + N	−0.0417	0.2906	0.6230	0.1109	3.47
	R	R + P + N	−0.3855	−0.0532	0.2791	0.9765	−0.63
	R + P	R + N	0.2132	0.5455	0.8779	0.0001	6.31
	R + P	R + P + N	−0.1306	0.2017	0.5340	0.4022	2.34
	R + N	R + P + N	−0.6762	−0.3438	−0.0115	0.0393	−4.25
During learning	R	R + P	−8.2838	−6.0490	−3.8143	0.0000	−3.45
	R	R + N	0.3106	2.5453	4.7801	0.0180	1.45
	R	R + P + N	−4.0134	−1.7786	0.4561	0.1717	−1.01
	R + P	R + N	6.3596	8.5944	10.8291	0.0000	4.74
	R + P	R + P + N	2.0356	4.2704	6.5051	0.0000	2.35
	R + N	R + P + N	−6.5587	−4.3240	−2.0892	0.0000	−2.50

Here, we analyze the average positioning speed and the cumulative number of steps needed during learning on the validation set of the best 4 hyperparameter sets for each condition. All hyperparameter sets were tested with 10 different randomly initialized networks. A positive percentage indicates the effect size in which Group 2 improves upon Group 1, whereas a negative percentage indicates the effect size in which Group 2 is worse than Group 1.