Table 4.

Training times for the NCFO architectures (seconds).

			Optimizer
(# of Users, # of Products)	# of Epochs	Batch Size	Adam	Adagrad	RMSprop
$(50,304)$	100	32	$145.59$	$189.61$	$141.35$
		64	$92.94$	$93.99$	$94.92$
		128	$62.35$	$63.99$	$63.37$
	200	32	$278.63$	$285.06$	$278.89$
		64	$184.44$	$190.17$	$181.16$
		128	$122.95$	$119.85$	$119.46$
	400	32	$566.35$	$555.67$	$565.06$
		64	$367.53$	$374.39$	$360.75$
		128	$240.70$	$236.37$	$236.96$
$(100,418)$	100	32	$250.34$	$249.61$	$568.37$
		64	$162.49$	$163.74$	$381.78$
		128	$109.51$	$106.37$	$245.87$
	200	32	$486.97$	$499.78$	$1158.23$
		64	$320.06$	$323.31$	$752.08$
		128	$211.67$	$205.29$	$483.98$
	400	32	$967.71$	$993.87$	$2750.17$
		64	$651.75$	$1518.80$	$2127.64$
		128	$403.53$	$949.84$	$397.96$
$(200,553)$	100	32	$482.37$	$484.87$	—
		64	$301.06$	$305.10$	—
		128	$198.08$	$189.73$	—
	200	32	$965.60$	$1213.02$	—
		64	$626.58$	$603.84$	—
		128	$402.97$	$389.71$	—
	400	32	$2005.65$	$1880.86$	—
		64	$1240.91$	$1177.66$	—
		128	$756.34$	$757.84$	—