search

udacity / robotics-nanodegree-issues

Public waffleboard to track Robotics Nanodegree Issues
2 stars 0 forks source link

Deep Learning: DNN lab: loss function #137

Closed royveshovda closed 7 years ago

royveshovda commented 7 years ago

Provided loss function return NaN. Ended up using loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits, labels=labels)) to get the lab to work.

sahiljuneja commented 7 years ago

That shouldn't happen though. But yes, the direct function you mention is better. I have updated the notebook. Thanks!