I think this part in model.py is wrong.

[SanoKyohei]

[Original]
for layer in range(self.n_layers):
weight_decay_loss = 0
weight_decay_loss += tf.multiply(tf.nn.l2_loss(self.W[layer]), self.weight_decay, name='weight_loss')
[Right]
weight_decay_loss = 0
for layer in range(self.n_layers):
weight_decay_loss += tf.multiply(tf.nn.l2_loss(self.W[layer]), self.weight_decay, name='weight_loss')

[RucDongLab]

Thank you very much for the correction! We fixed this error in the new code. Due to the relatively shallow network structure, this error did not have a big impact on the results.