search

fferroni / DEC-Keras

Deep Embedding Clustering in Keras
GNU General Public License v3.0
129 stars 59 forks source link

loss: nan #9

Open antoniosap opened 6 years ago

antoniosap commented 6 years ago

HI,

i have this value. Any ideas?

dipanjannag commented 6 years ago

Facing same issue with simple code

c = DeepEmbeddingClustering(n_clusters=10, input_dim=17)
c.initialize(X, finetune_iters=100, layerwise_pretrain_iters=500)
c.cluster(X, y=labels)

the loss is quickly reaching nan even before one full iteration

ganaz commented 6 years ago

check if any column has all zero value and remove them , the another thing you can do is reduce learning rate. I think by reducing learning rate may fix this issue.