遇到的坑及其解决方案

1）graphviz.backend.ExecutableNotFound: failed to execute ['dot', '-Tpdf', '-O', 'plot.gv'], make sure the Graphviz executables are on your systems' PATH

解决方案：安装graphviz

https://blog.csdn.net/robin_xu_shuai/article/details/53355711

2）R - mxnet RNN “Incompatible attr in node at 0-th output”

问题可能原因：

类别数目不对；

3）OOM when allocating tensor with shape[128,64,112,112]

或 src/storage/./pooled_storage_manager.h:143: cudaMalloc failed: out of memory

训练时：显卡内存不够，可能原因：

batch_size太大，内存不够；

显卡内存被别的程序占用没有退出，可在后台查看占用线程并终止。