训练Mnist CNN 网络
根据TensorFlow 提供的Offcial/mnist提供的代码,第一次未能正确运行,修改个人目录下的.bashrc文件,添加一句:
export CUDA_VISIBLE_DEVICES="1"
之后source ~/.bashrc
然后运行:python mnist.py 会自动调用GPU运行训练网络。
启动Tensorboard观察训练曲线:
tensorboard --logdir mnist_cnn_model
之后在浏览器中打开: http://zc-GE62-2QD:6006,(链接因人而异)即可调出训练曲线和Graph。
重启电脑之后无法失败nvidia-smi命令
重启电脑之后电脑无法失败之前安装的Nvidia 驱动,按照Install blog的介绍,加入如下配置:
1、检查BIOS启动项,关闭一些选项
在开机启动项的Security选项中检查UEFI是否开启,如果开启的话请立马关掉它(重要)
在开机启动项的Boot选项中检查Secure Boot是否开启,如果开启的话请立马关掉它(重要)
2、安装相关依赖
sudo apt-get install libprotobuf-dev libleveldb-dev libsnappy-dev libopencv-dev libhdf5-serial-dev protobuf-compiler
sudo apt-get install --no-install-recommends libboost-all-dev
sudo apt-get install libopenblas-dev liblapack-dev libatlas-base-dev
sudo apt-get install libgflags-dev libgoogle-glog-dev liblmdb-dev
如果有图形化界面则需要禁用x-window服务
sudo service lightdm stop
或
sudo /etc/init.d/lightdm stop
3、禁用 nouveau
安装好依赖包后需要禁用 nouveau,只有在禁用掉 nouveau 后才能顺利安装 NVIDIA 显卡驱动,禁用方法就是在 sudo vim /etc/modprobe.d/blacklist.conf文件中添加一条禁用命令
blacklist nouveau
blacklist lbm-nouveau
options nouveau modeset=0
alias nouveau off
alias lbm-nouveau off
执行如下命令:
echo options nouveau modeset=0 | sudo tee -a /etc/modprobe.d/nouveau-kms.conf build the new kernel by:
最后更新后重启电脑
sudo update-initramfs -u
reboot
之后按照之前的博客invidia 驱动安装
命令行无法识别GPU的问题
按照上述的步骤完成之后,可以在PyCharm 的环境下运行mnist.py文件使用GPU进行训练,但是在命令行里调用该只来会提示:
2018-08-20 00:06:22.191504: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
2018-08-20 00:06:22.337492: E tensorflow/stream_executor/cuda/cuda_driver.cc:406] failed call to cuInit: CUDA_ERROR_NO_DEVICE
2018-08-20 00:06:22.337549: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:158] retrieving CUDA diagnostic information for host: zc-GE62-2QD
2018-08-20 00:06:22.337558: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:165] hostname: zc-GE62-2QD
2018-08-20 00:06:22.337611: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:189] libcuda reported version is: 396.26.0
2018-08-20 00:06:22.337639: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:193] kernel reported version is: 396.26.0
2018-08-20 00:06:22.337647: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:300] kernel version seems to match DSO: 396.26.0
0
I0820 00:06:22.339188 140026981222144 tf_logging.py:116] Using default config.
I0820 00:06:22.339440 140026981222144 tf_logging.py:116] Using config: {'_model_dir': './mnist_cnn_model', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_session_config': None, '_keep_checkpoint_max': 5, '_keep_checkpoint_every_n_hours': 10000, '_log_step_count_steps': 100, '_train_distribute': None, '_service': None, '_cluster_spec': <tensorflow.python.training.server_lib.ClusterSpec object at 0x7f5a2f8b9278>, '_task_type': 'worker', '_task_id': 0, '_global_id_in_cluster': 0, '_master': '', '_evaluation_master': '', '_is_chief': True, '_num_ps_replicas': 0, '_num_worker_replicas': 1}
I0820 00:06:22.387485 140026981222144 tf_logging.py:116] Calling model_fn.
I0820 00:06:22.727335 140026981222144 tf_logging.py:116] Done calling model_fn.
I0820 00:06:22.728235 140026981222144 tf_logging.py:116] Create CheckpointSaverHook.
I0820 00:06:22.891083 140026981222144 tf_logging.py:116] Graph was finalized.
I0820 00:06:22.892676 140026981222144 tf_logging.py:116] Restoring parameters from ./mnist_cnn_model/model.ckpt-62405
I0820 00:06:22.968533 140026981222144 tf_logging.py:116] Running local_init_op.
I0820 00:06:22.973660 140026981222144 tf_logging.py:116] Done running local_init_op.
2018-08-20 00:06:23.323585: E tensorflow/core/common_runtime/executor.cc:660] Executor failed to create kernel. Invalid argument: Default MaxPoolingOp only supports NHWC on device type CPU
解决办法:
在命令行丽执行:
CUDA_VISIBLE_DEVICES=0
python mnist.py
输出:
正常的训练log
(tensorflow) zc@zc-GE62-2QD:~/workspace/go_elife/AI_Demo/mnist_cnn$ python mnist.py
2018-08-20 00:27:17.136794: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
2018-08-20 00:27:17.400050: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:898] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2018-08-20 00:27:17.400468: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1356] Found device 0 with properties:
name: GeForce GTX 960M major: 5 minor: 0 memoryClockRate(GHz): 1.176
pciBusID: 0000:01:00.0
totalMemory: 1.96GiB freeMemory: 1.92GiB
2018-08-20 00:27:17.400488: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1435] Adding visible gpu devices: 0
2018-08-20 00:27:29.237605: I tensorflow/core/common_runtime/gpu/gpu_device.cc:923] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-08-20 00:27:29.237665: I tensorflow/core/common_runtime/gpu/gpu_device.cc:929] 0
2018-08-20 00:27:29.237684: I tensorflow/core/common_runtime/gpu/gpu_device.cc:942] 0: N
2018-08-20 00:27:29.237926: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1053] Created TensorFlow device (/device:GPU:0 with 1687 MB memory) -> physical GPU (device: 0, name: GeForce GTX 960M, pci bus id: 0000:01:00.0, compute capability: 5.0)
1
I0820 00:27:29.270864 140173382854400 tf_logging.py:116] Using default config.
I0820 00:27:29.271144 140173382854400 tf_logging.py:116] Using config: {‘_model_dir’: ‘./mnist_cnn_model’, ‘_tf_random_seed’: None, ‘_save_summary_steps’: 100, ‘_save_checkpoints_steps’: None, ‘_save_checkpoints_secs’: 600, ‘_session_config’: None, ‘_keep_checkpoint_max’: 5, ‘_keep_checkpoint_every_n_hours’: 10000, ‘_log_step_count_steps’: 100, ‘_train_distribute’: None, ‘_service’: None, ‘_cluster_spec’:
调用Model运行测试
保留训练的文件到saved_model,关于: SavedModel
python mnist.py --export_dir ./mnist_cnn_saved_model
上述命令hi执行一圈训练,并把训练的Model保存。
``
saved_model_cli run --dir ./mnist_cnn_saved_model/1534951616/ --tag_set serve --signature_def classify --inputs image=examples.npy
输出:
2018-08-22 23:48:29.194109: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
2018-08-22 23:48:29.396440: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:898] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2018-08-22 23:48:29.396802: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1356] Found device 0 with properties:
name: GeForce GTX 960M major: 5 minor: 0 memoryClockRate(GHz): 1.176
pciBusID: 0000:01:00.0
totalMemory: 1.96GiB freeMemory: 1.92GiB
2018-08-22 23:48:29.396817: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1435] Adding visible gpu devices: 0
2018-08-22 23:48:34.662749: I tensorflow/core/common_runtime/gpu/gpu_device.cc:923] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-08-22 23:48:34.662780: I tensorflow/core/common_runtime/gpu/gpu_device.cc:929] 0
2018-08-22 23:48:34.662792: I tensorflow/core/common_runtime/gpu/gpu_device.cc:942] 0: N
2018-08-22 23:48:34.662945: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1053] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 1687 MB memory) -> physical GPU (device: 0, name: GeForce GTX 960M, pci bus id: 0000:01:00.0, compute capability: 5.0)
Result for output key classes:
[3 3]
Result for output key probabilities:
[[2.8152197e-18 5.6790460e-20 6.9913247e-18 1.0000000e+00 0.0000000e+00
6.2972307e-09 2.4830664e-19 3.2860752e-29 1.6266099e-26 5.6905280e-09]
[5.3256835e-13 1.2014016e-10 2.4255571e-09 9.9993646e-01 3.9114351e-15
4.1689669e-05 1.8050653e-07 6.5699896e-23 5.9275825e-19 2.1656551e-05]]
输出模型预测的结果。上述结果出错,应该是训练迭代次数过多,出现了过拟合的现象,重新训练了一次模型:预测结果如下:
(tensorflow) zc@zc-GE62-2QD:~/workspace/go_elife/AI_Demo/mnist_cnn$ saved_model_cli run --dir ./mnist_cnn_saved_model/1534953492/ --tag_set serve --signature_def classify --inputs image=examples.npy
2018-08-22 23:59:09.114953: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
2018-08-22 23:59:09.317977: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:898] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2018-08-22 23:59:09.318344: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1356] Found device 0 with properties:
name: GeForce GTX 960M major: 5 minor: 0 memoryClockRate(GHz): 1.176
pciBusID: 0000:01:00.0
totalMemory: 1.96GiB freeMemory: 1.92GiB
2018-08-22 23:59:09.318364: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1435] Adding visible gpu devices: 0
2018-08-22 23:59:09.828057: I tensorflow/core/common_runtime/gpu/gpu_device.cc:923] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-08-22 23:59:09.828108: I tensorflow/core/common_runtime/gpu/gpu_device.cc:929] 0
2018-08-22 23:59:09.828121: I tensorflow/core/common_runtime/gpu/gpu_device.cc:942] 0: N
2018-08-22 23:59:09.828241: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1053] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 1687 MB memory) -> physical GPU (device: 0, name: GeForce GTX 960M, pci bus id: 0000:01:00.0, compute capability: 5.0)
Result for output key classes:
[5 3]
Result for output key probabilities:
[[1.4278330e-03 4.7121157e-10 5.4629195e-06 4.7545331e-03 2.8239561e-12
5.6043041e-01 4.3334922e-01 6.1847216e-12 3.2551703e-05 1.2838069e-08]
[2.0283554e-02 7.5934164e-05 1.5059297e-02 7.2744995e-01 8.3420676e-04
2.1087909e-01 2.5073985e-02 3.2057675e-12 3.4292994e-04 9.9194108e-07]]
调用图片预测
调用图片读取需要用到Oppencv,直接安装conda install opencv
conda install opencv
不推荐用此指令安装,我用此指令安装把我的Tensorflow版本降低为1.5.0,气人。
tensorflow版本切换:
conda install --channel https://conda.anaconda.org/anaconda tensorflow-gpu=1.8.0
再次遇到:NVIDIA-SMI has failed because it couldn’t communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.
运行命令:sudo apt-get install cuda-drivers
nvidia-smi
暂时解决问题,感觉不是从根本上解决的问题。
- Markdown和扩展Markdown简洁的语法
- 代码块高亮
- 图片链接和图片上传
- LaTex数学公式
- UML序列图和流程图
- 离线写博客
- 导入导出Markdown文件
- 丰富的快捷键
快捷键
- 加粗
Ctrl + B
- 斜体
Ctrl + I
- 引用
Ctrl + Q
- 插入链接
Ctrl + L
- 插入代码
Ctrl + K
- 插入图片
Ctrl + G
- 提升标题
Ctrl + H
- 有序列表
Ctrl + O
- 无序列表
Ctrl + U
- 横线
Ctrl + R
- 撤销
Ctrl + Z
- 重做
Ctrl + Y
Markdown及扩展
Markdown 是一种轻量级标记语言,它允许人们使用易读易写的纯文本格式编写文档,然后转换成格式丰富的HTML页面。 —— [ 维基百科 ]
使用简单的符号标识不同的标题,将某些文字标记为粗体或者斜体,创建一个链接等,详细语法参考帮助?。
本编辑器支持 Markdown Extra , 扩展了很多好用的功能。具体请参考Github.
表格
Markdown Extra 表格语法:
项目 | 价格 |
---|---|
Computer | $1600 |
Phone | $12 |
Pipe | $1 |
可以使用冒号来定义对齐方式:
项目 | 价格 | 数量 |
---|---|---|
Computer | 1600 元 | 5 |
Phone | 12 元 | 12 |
Pipe | 1 元 | 234 |
定义列表
-
Markdown Extra 定义列表语法:
项目1
项目2
- 定义 A
- 定义 B 项目3
- 定义 C
-
定义 D
定义D内容
代码块
代码块语法遵循标准markdown代码,例如:
@requires_authorization
def somefunc(param1='', param2=0):
'''A docstring'''
if param1 > param2: # interesting
print 'Greater'
return (param2 - param1 + 1) or None
class SomeClass:
pass
>>> message = '''interpreter
... prompt'''
脚注
生成一个脚注1.
目录
用 [TOC]
来生成目录:
数学公式
使用MathJax渲染LaTex 数学公式,详见math.stackexchange.com.
- 行内公式,数学公式为: Γ(n)=(n−1)!∀n∈ℕ Γ ( n ) = ( n − 1 ) ! ∀ n ∈ N 。
- 块级公式:
更多LaTex语法请参考 这儿.
UML 图:
可以渲染序列图:
或者流程图:
离线写博客
即使用户在没有网络的情况下,也可以通过本编辑器离线写博客(直接在曾经使用过的浏览器中输入write.blog.youkuaiyun.com/mdeditor即可。Markdown编辑器使用浏览器离线存储将内容保存在本地。
用户写博客的过程中,内容实时保存在浏览器缓存中,在用户关闭浏览器或者其它异常情况下,内容不会丢失。用户再次打开浏览器时,会显示上次用户正在编辑的没有发表的内容。
博客发表后,本地缓存将被删除。
用户可以选择 把正在写的博客保存到服务器草稿箱,即使换浏览器或者清除缓存,内容也不会丢失。
注意:虽然浏览器存储大部分时候都比较可靠,但为了您的数据安全,在联网后,请务必及时发表或者保存到服务器草稿箱。
浏览器兼容
- 目前,本编辑器对Chrome浏览器支持最为完整。建议大家使用较新版本的Chrome。
- IE9以下不支持
- IE9,10,11存在以下问题
- 不支持离线功能
- IE9不支持文件导入导出
- IE10不支持拖拽文件导入
- 这里是 脚注 的 内容. ↩