the ntm model loss is very large

[zhunipingan]

hi, thanks for your sharing. the code is high-quality and clear. But when I attempt to use your model with my own dataset, I find the ntm_loss is very large. I don't know why, can you give me some help. the bow_dictionary_vocab_size is set to 10000.

09/14/2021 11:22:43 [INFO] train_mixture: ====>Train epoch: 109 Average loss: 6429827.4577 09/14/2021 11:22:43 [INFO] train_mixture: Overall sparsity = 0.999, l1 strength = 81.71844 09/14/2021 11:22:43 [INFO] train_mixture: Target sparsity = 0.850 09/14/2021 11:22:44 [INFO] train_mixture: ====> Test epoch: 109 Average loss: 6389828.5666 09/14/2021 11:22:51 [INFO] train_mixture: ====>Train epoch: 110 Average loss: 9138395.1334 09/14/2021 11:22:51 [INFO] train_mixture: Overall sparsity = 0.999, l1 strength = 73.68993 09/14/2021 11:22:51 [INFO] train_mixture: Target sparsity = 0.850 09/14/2021 11:22:52 [INFO] train_mixture: ====> Test epoch: 110 Average loss: 7285204.7357 09/14/2021 11:22:52 [INFO] train_mixture: 09/14/2021 11:23:01 [INFO] train_mixture: ====>Train epoch: 111 Average loss: 4194444.3072 09/14/2021 11:23:01 [INFO] train_mixture: Overall sparsity = 0.999, l1 strength = 66.44983 09/14/2021 11:23:01 [INFO] train_mixture: Target sparsity = 0.850 09/14/2021 11:23:01 [INFO] train_mixture: ====> Test epoch: 111 Average loss: 15234667.0275