Question about initialize mentornet epoch embedding

google / mentornet

Code for MentorNet: Learning Data-Driven Curriculum for Very Deep Neural Networks

Apache License 2.0

320 stars 63 forks source link

Question about initialize mentornet epoch embedding #4

Closed ironschen closed 4 years ago

ironschen commented 5 years ago

epoch_embedding = tf.get_variable( 'epoch_embedding', [100, epoch_embedding_size], trainable=False) Because epoch_embedding is not tainable, how to initialize this variable when train mentornet from zero?

roadjiang commented 5 years ago

I used the default initializer. We have also tried trainable=False and did not find big difference in performance.