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

A question about micro child model #48

Closed bl0 closed 6 years ago

bl0 commented 6 years ago

In the following line, why use layers = [layers[-1], x] instead of layers = [layers[0], x]?

https://github.com/melodyguan/enas/blob/d1a90ac915301198f2a30ce136e9040e6c4235ff/src/cifar10/micro_child.py#L277

In my opinion, if use your implementation, the model in the search stage is very different from that in the final stage.

MarkAlive commented 6 years ago

I changed [layers[-1], x] to [layers[0], x] and get a better result.