NAO_pytorch

NAO_pytorch copied to clipboard

OSError: [Errno 39] Directory not empty: '/tmp/pymp-yffyhr_a'

4

When I run the script "train_search_cifar10.sh", I got the following error message: Is there anyone meet the same problem? 06/26 04:36:37 AM epoch 200 lr 1.001480e-03 06/26 04:37:52 AM Train...

HKLee2040

hey dear： if i set child_layers=1, child_nodes=2, child_channels=48, child_num_ops=5, due to args.controller_encoder_vocab_size = 1 + ( args.child_nodes + 2 - 1 ) + args.child_num_ops=9 cant fit NAO encoder model.....

lucaskyle

代码没看懂可以寻求帮助吗╥﹏╥

3

您好！我想学习一下您的工作，但是我没有看懂您的代码。 train_search.py里面定义了一个[model](https://github.com/renqianluo/NAO_pytorch/blob/master/train_search.py#L425)网络结构。这是要搜索的网络结构吗？ 在427行定义了一个[nao](https://github.com/renqianluo/NAO_pytorch/blob/master/train_search.py#L427)这里面是包含了编码器和解码器吗。这个用来求解出最好的结构？ 我没有明白，nao会通过一个[num_new_archs](https://github.com/renqianluo/NAO_pytorch/blob/master/train_search.py#L536)变量改变model，但是model怎么影响nao呢？nao怎么知道model这个模型设计的好不好

zihaozhang9

I could be wrong, but shouldn't there be a ReLU before the global pooling in NASNetworkCIFAR and NASNetworkImageNet? The tensorflow version of this code seems to have one: https://github.com/renqianluo/NAO/blob/master/NAO/cnn/model.py#L682

fps7806