Child Hyper Parameters - Githubissues

melodyguan / enas

TensorFlow Code for paper "Efficient Neural Architecture Search via Parameter Sharing"

https://arxiv.org/abs/1802.03268

Apache License 2.0

1.58k stars 390 forks source link

Open ghost opened 6 years ago

ghost commented 6 years ago

Hi, could you briefly explain the meaning of: "child_out_filters", "child_num_branches", "child_num_cell_layers", and "child_keep_prob"

I am fairly new to ML and am trying to figure out how to properly train the outputted fixed architectures.

Thank you so much! This really is a wonderful tool.

karandwivedi42 commented 6 years ago

"child_out_filters" -> Number of channels of the convolution layers

"child_num_branches", -> Number of options in ENAS (here it is fixed at 5)

"child_num_cell_layers" -> Can't find this in the code..

"child_keep_prob" -> dropout keep probability more here