Input layer dimensions are not set if there is no training data reader

LLNL / lbann

Livermore Big Artificial Neural Network Toolkit

http://software.llnl.gov/lbann/

Other

222 stars 78 forks source link

Input layer dimensions are not set if there is no training data reader #1098

Open timmoon10 opened 5 years ago

timmoon10 commented 5 years ago

I'd like to run models without a training set. In particular, unit tests with the "check metric" and "check gradient" callbacks only need the test set. However, if I don't specify a training data reader, the input layer is initialized with output dimensions of 0. I can workaround this for now by setting up a dummy data reader for training and setting the number of epochs to zero.

samadejacobs commented 5 years ago

I think this is a problem with "inference-only LBANN" too.