Table 3.

Summary of generalization experiments to test datasets with longer planning horizons

LSTM Train			Test Data			pred	timeCPX	timeML	timeimp	timegain	inf	optgap
c	f	T	c	f	T	(%)				(%)	(%)	(%)
3	1000	90	3	1000	180	25	0.5	0.5	1	7.6	0.0	0.1
						50		0.4	1	20.5	0.0	0.1
						75		0.4	1	29.1	0.0	0.2
						85		0.4	1	33.3	0.0	0.2
						90		0.3	2	38.2	0.0	0.2
						95		0.3	2	44.2	0.1	0.2
						100		0.1	6	82.4	1.0	0.4
						100(UC)		0.3	2	36.2	0.0	0.4
3	1000	90	3	1000	360	25	1.2	1.0	1	19.7	0.0	0.1
						50		0.7	2	41.6	0.0	0.2
						75		0.5	2	59.6	0.0	0.2
						85		0.4	3	64.2	0.0	0.3
						90		0.4	3	66.4	0.0	0.3
						95		0.3	3	71.4	0.2	0.3
						100		0.1	12	91.9	1.3	0.5
						100(UC)		0.4	3	62.8	0.0	0.5
5	10000	90	5	10000	180	25	19.0	6.2	3	67.2	0.0	0.3
						50		2.2	9	88.4	0.0	0.5
						75		0.7	28	96.4	0.1	0.5
						85		0.4	46	97.8	0.3	0.6
						90		0.3	59	98.3	0.9	0.7
						95		0.2	78	98.7	2.8	0.9
						100		0.1	166	99.4	27.5	3.5
						100(UC)		0.4	49	98.0	0.0	1.6
8	10000	120	8	10000	480	25	42.4	11.6	4	72.7	0.0	0.6
						50		4.3	10	89.9	0.0	0.9
						75		1.7	25	96.1	0.0	0.9
						85		0.8	55	98.2	0.0	0.9
						90		0.5	86	98.8	0.1	1.0
						95		0.4	112	99.1	0.4	1.0
						100		0.1	364	99.7	5.2	2.8
						100(UC)		0.6	71	98.6	0.0	1.5

^*Experiments include 20000 test instances