由于最近没时间,花了一天搭了一下,个人总结了点小tips
Markdown 直接粘过来的,
包含了Vmware+Ubuntu搭配虚拟机(目前没用到
个人配置 Win10 + VS2017 + CUDA9 + cuDNN7 + Py3.5 + tensorflow1.5 + keras2.1.4 + opencv最新 + Anaconda最新 可以正常使用。
安装顺序 VS2017 -- Anaconda --- Python --- CUDA --- cuDNN --- Tensorflow-GPU --- PyCharm
# 教程
http://www.bubuko.com/infodetail-2465293.html
https://blog.csdn.net/XunCiy/article/details/89016510
https://www.cnblogs.com/caizhou520/p/11219985.html
https://www.cnblogs.com/yuxuefeng/articles/9235431.html
## 报错
#### 1type
https://blog.csdn.net/bigdream123/article/details/99467316
_np_qint8 = np.dtype([("quint8", np.uint8, (1,))])
_np_quint8 = np.dtype([("quint8", np.uint8, (1,))])
_np_qint16 = np.dtype([("qint16", np.int16, (1,))])
_np_quint16 = np.dtype([("quint16", np.uint16, (1,))])
_np_qint32 = np.dtype([("qint32", np.int32, (1,))])
np_resource = np.dtype([("resource", np.ubyte, (1,))])
#### Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 (CPU垃圾)
https://blog.csdn.net/jackfjw/article/details/83046283
#### failed to create cublas handle: CUBLAS_STATUS_ALLOC_FAILED (内存不够)
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.333)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options))
## pycharm
https://www.cnblogs.com/wqzn/p/10424892.html
## VMWARE
https://www.7down.com/soft/310739.html
* 开机 F2-config-Visual Tech 开启
## Ubuntu
https://www.jianshu.com/p/94aa39bcd39d?tdsourcetag=s_pcqq_aiomsg
## pip镜像
1) C:\users\xxx\ 含有.python 什么的文件夹 新建文件夹pip
2) 新建文件pip.ini并编辑如下
[global]
index-url = https://pypi.tuna.tsinghua.edu.cn/simple
[install]
trusted-host=mirrors.aliyun.com
#### GPU-Tensorflow 需要CUDA,cuDNN
## CUDA & cuDNN
1) 计算机-管理-系统工具-设备管理器-显示适配器查-看显卡
2) 控制面板-NVIDIA控制面板-帮助-系统信息-组件-NVCUDA.DLL 看最高支持版本
3) “C:\Program Files\NVIDIA Corporation\NVSMI” --- 路径加入计算机的Path
cmd -- nvidia-smi -- 查看Driver Version 查看最高支持版本
* CUDA-cuDNN--版本搭配
* 下载CUDA & cuDNN
https://www.cnblogs.com/xiaojianliu/p/9286066.html
cuDNN解压后文件移动到同名CUDA文件夹内
* 验证CUDA版本
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\extras\demo_suite\
bandwidthTest.exe & deviceQuery.exe cmd运行是否pass
## VS2017
https://www.jianshu.com/p/320aefbc582d
## Anaconda
https://www.anaconda.com/distribution/
* 安装时候add path
* 安装完cmd改conda镜像
conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
conda config --set show_channel_urls yes
#### 基本命令
* 更新包
//conda upgrade --all
* 装np,注意:每个环境都要重新装一遍包,tensorflow不要装会冲突
conda install numpy scipy pandas
* 查看所有包配置(pip freeze)
conda env export
* 查看所有环境
conda env list
* 退出环境
conda deactivate tensorflow_py3.5
* 删除环境
conda remove -n tensorflow_py3.5 --all
* 创建环境
conda create --name tensorflow_py3.5 python=3.5
* 激活环境
conda activate tensorflow_py3.5
* 复制环境
conda create -n dltest_py3.5 --clone deeplearning_py3.5
#### 安装
* 创建环境
conda create --name tensorflow_py3.5 python=3.5
* 激活环境
conda activate tensorflow_py3.5
* 升级pip
python -m pip install --upgrade pip
* 最新版tf,慎用
* pip3 install --ignore-installed --upgrade tensorflow-gpu
* 安装tf-GPU版本
pip3 install tensorflow-gpu==1.5
* 安装tf-CPU
pip3 install --ignore-installed --upgrade tensorflow
* 配置CUDA适应tf
dlerror: cudart64_100.dll not found C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin\ -- 相应文件后缀改成100
##### tensorflow测试
import tensorflow as tf
#构造计算图
hello = tf.constant("Hello")
#执行计算图
sess = tf.Session()
print(sess.run(hello))
##### Git
conda install git
##### Keras
pip3 install -U keras=2.1.4
* 测试
https://blog.csdn.net/Snowy_susu/article/details/81836824
##### OpenCV
pip3 install -U opencv-contrib-python
* 测试
import cv2
import numpy as np
img = cv2.imread('D:/aa.jpg', cv2.IMREAD_COLOR)
cv2.imshow("image", img)
cv2.waitKey(0)