Table 2.

The comparison results of popular algorithms.

	Adam	Rmsprop	Adagrad	Momentum	Gradient Descent
Initial learning rate	0.0001
Epoch number	1000
Clustering loss	0.007	0.012	0.099	0.061	0.130
Classify loss	0.314	0.314	0.694	0.693	0.694
Train data accuracy	100.00%	99.91%	10.13%	50.00%	50.00%
Val. data accuracy	99.61%	99.65%	12.11%	50.00%	50.00%
Test data accuracy	91.29%	50.00%	50.00%	50.00%	50.00%