最新消息:

Turbot-DL入门教程篇-深度学习框架TensorFlow安装

Turtlebot2/二代机器人 少儿编程 1395浏览 0评论
Turbot-DL入门教程

Turbot-DL入门教程篇-深度学习框架TensorFlow安装

说明:

  • 介绍如何在turbot-DL上安装深度学习框架TensorFlow

准备:

  • 利用Jetpack安装如下:

    • L4T 24.2.1 an Ubuntu 16.04 64-bit variant (aarch64)
    • CUDA 8.0
    • cuDNN 5.1.5
  • TensorFlow安装需要用到CUDA和cuDNN

  • TensorFlow占用比较多空间,TX1通常空间不足,最好增加64G+的U盘作为root分区启动,增加交换分区大小为8G+

  • 或者挂载U盘作为分区,移植大文件或目录到U盘,并做好ln软连接到原来的目录位置。

  • 总体测试下来,安装时间比较长,会有错误出现,根据错误做修复。

  • 如果入口脚本命令报错的话,请根据脚本里面的命令一条条执行,并处理错误信息。并可以注释下载类命令,再重复执行。

安装:

  • 下载脚本:
mkdir ~/dl
cd ~/dl
git clone https://github.com/jetsonhacks/installTensorFlowTX1.git
cd installTensorFlowTX1
  • 使用/usr/local/lib库
$ ./setLocalLib.sh
  • 安装依赖:
chmod +x installPrerequisites.sh
./installPrerequisites.sh
  • 包含安装java,Protobuf,grpc-java ,Bazel等
  • 下载tensorflow代码
$ ./cloneTensorFlow.sh
  • 设置环境变量
$ ./setTensorFlowEV.sh
  • 编译TensorFlow
$ ./buildTensorFlow.sh
  • 编译时间比较长,中间会有自动退出的情况,继续执行命令直到完成
  • 打包成whl文件,放在$HOME目录下,如:tensorflow-0.11.0-py2-none-any.whl
$ ./packageTensorFlow.sh
  • 安装whl文件
$ pip install $HOME/tensorflow-0.11.0-py2-none-any.whl
  • 提示权限问题的话,pip前加sudo

测试:

  • 运行TensorFlow例子:
$ cd $HOME/tensorflow
$ time python tensorflow/models/image/mnist/convolutional.py
  • 效果如下:
Minibatch loss: 1.662, learning rate: 0.006302
Minibatch error: 0.0%
Validation error: 0.8%
Step 7900 (epoch 9.19), 54.4 ms
Minibatch loss: 1.647, learning rate: 0.006302
Minibatch error: 0.0%
Validation error: 0.8%
Step 8000 (epoch 9.31), 54.6 ms
Minibatch loss: 1.671, learning rate: 0.006302
Minibatch error: 1.6%
Validation error: 0.9%
Step 8100 (epoch 9.43), 54.6 ms
Minibatch loss: 1.625, learning rate: 0.006302
Minibatch error: 0.0%
Validation error: 0.8%
Step 8200 (epoch 9.54), 54.2 ms
Minibatch loss: 1.633, learning rate: 0.006302
Minibatch error: 0.0%
Validation error: 0.9%
Step 8300 (epoch 9.66), 54.4 ms
Minibatch loss: 1.616, learning rate: 0.006302
Minibatch error: 0.0%
Validation error: 0.8%
Step 8400 (epoch 9.77), 54.5 ms
Minibatch loss: 1.604, learning rate: 0.006302
Minibatch error: 0.0%
Validation error: 0.7%
Step 8500 (epoch 9.89), 54.0 ms
Minibatch loss: 1.604, learning rate: 0.006302
Minibatch error: 1.6%
Validation error: 0.9%
Test error: 0.8%

real    8m34.741s
user    4m11.700s
sys         1m24.880s

问题集:

  • 问题1:提示找不到http://zlib.net/zlib-1.2.8.tar.gz
  • 解决:需要打补丁,cloneTensorFlow.sh里面的打补丁没执行好
  • 问题2:提示junit.4.12.jar不能下载
  • 解决:修改vim ~/tensorflow/tensorflow/workspace.bzl 更换地址,并去掉校验
  native.http_jar(
      name = "junit_jar",
      #url = "https://github.com/junit-team/junit4/releases/download/r4.12/junit-4.12.jar",
      url = "https://github.com/orrsella/bazel-example/blob/master/third_party/junit/junit-4.12.jar",
      #sha256 = "59721f0805e223d84b90677887d9ff567dc534d7c502ca903c0c2b17f05c116a",
      #sha256 = "fe3d4c56388dc3d74049abae83f4520f6703062e174e16bb5551cdf439ca4f81",
  )
  • 问题3:可能会出现多次卡死在编译阶段:./buildTensorFlow.sh
  • 解决:增加swap空间为4G,参考方法

参考:

  • https://www.tensorflow.org/get_started/os_setup.html
  • https://devtalk.nvidia.com/default/topic/901148/tensorflow-on-jetson-tx1/?offset=11
  • https://github.com/tensorflow/tensorflow/
  • https://github.com/tensorflow/tensorflow/issues/851
  • http://www.jetsonhacks.com/2016/12/30/tensorflow-nvidia-jetson-tx1-development-kit/
  • https://github.com/jetsonhacks/installTensorFlowTX1
  • http://www.jetsonhacks.com/2016/12/30/install-tensorflow-on-nvidia-jetson-tx1-development-kit/
  • http://www.jetsonhacks.com/2017/01/15/tensorflow-build-update-jetson-tx1/

您必须 登录 才能发表评论!