A Reweighted Scheme to Improve the Representation of the Neural Autoregressive Distribution Estimator

Table 1

Details about the twelve datasets.

Dataset	Input size	Number of classes	Training	Validation	Testing
Adult	123	2	5000	1414	26147
Binarized-MNIST	784	10	50000	10000	10000
Connect-4	126	3	16000	4000	47557
Convex	784	2	6000	2000	50000
DNA	180	3	1400	600	1186
Mushrooms	112	2	2000	500	5624
Newsgroups	5000	20	9578	1691	7505
OCR-letters	128	26	32152	10000	10000
RCV1	150	2	40000	10000	150000
Rectangles	784	2	1000	200	50000
SVHN	1024	11	594388	10000	26032
Web	300	2	14000	3188	32561