Not able to reproduce results

[gullalc]

I am trying to reproduce the results on all the datasets given in the paper. I am using the default parameters in the code, but strangely the loss for autoencoder training keeps on increasing and eventually reaches nan. See below the loss for 1st 5 epochs.

print('Epoch: ', epoch, '| reconst ae loss:%f ' % (tot_loss.data[0])) ('Epoch: ', 0, '| reconst ae loss:119277.914062 ') ('Epoch: ', 1, '| reconst ae loss:1317753.750000 ') ('Epoch: ', 2, '| reconst ae loss:19503726.000000 ') ('Epoch: ', 3, '| reconst ae loss:304296928.000000 ') ('Epoch: ', 4, '| reconst ae loss:5205163008.000000 ') ('Epoch: ', 5, '| reconst ae loss:94487601152.000000 ')

[shashitn]

Same here. Any advice?

[balloch]

Any updates on this?