enas
enas copied to clipboard
TensorFlow Code for paper "Efficient Neural Architecture Search via Parameter Sharing"
I want to use ENAS arch to inference my own dataset, so I modified the original code to use placeholder input e.g) ``` loss, lr, gn, tr_acc, _ = sess.run(run_ops,...
Here is my question: after training, what should be the output of the nas? Is that the best one ever during the training or the last architecture produced by the...
When I try to run the cifar10_mirco_search.sh , some errors happened: File "/home/zhangjiabin/enas/src/cifar10/micro_child.py", line 602, in _enas_conv zero_init = tf.initializers.zeros(dtype=tf.float32) AttributeError: 'module' object has no attribute 'initializers' It seems caused...
What is the mean about the “whole_channel is True”? Looking forward to your reply!Thanks
Based on the output of micro_search(`num_cell=2`), for example: `[0 0 1 0 0 4 2 0]` `[0 3 0 1 1 4 0 1]` According to the explanation, each block...
If we run the three experiments from the README: ``` # Exp. 1 ./scripts/ptb_search.sh ./scripts/ptb_final.sh # Exp. 2 ./scripts/cifar10_macro_search.sh ./scripts/cifar10_macro_final.sh # Exp 3. ./scripts/cifar10_micro_search.sh ./scripts/cifar10_micro_final.sh ``` what should we expect...
How does the child_fixed_arc are created on the final scripts Is that use the controller trained on the search phase to output the child_fixed_arc?
Resolves https://github.com/melodyguan/enas/issues/4
Thank you for providing us with the code. I'm running the cifar10_macro_search.sh The code, data, hyper parameter were taken as is and not modified. tail of cifar10_macro_search.sh(in the stdout file)...
I have read the enas paper,which says that weight sharing is used for efficiency. Any one can tell me which part of the code realize the weight sharing?