tensorflow-gpu1.14.0/cuda10.0/cudnn7.4.1环境搭建

​ 我的个人电脑是win10系统，其GPU是双显卡：1.Inter® UHD Graphics 630；2.NVIDIA GeForce GTX 1660Ti，显存为6GB，因此我去搭建tensorflow-gpu环境。

搭建好的环境：

• win10系统
• pycharm与Anaconda3-2019.10软件
• 在Anaconda3-2019.10下创建虚拟环境DeepLearnFromTensorflow
• 在DeepLearnFromTensorflow下安装python=3.7、tensorflow-gpu1.14.0、cuda 10.0, V10.0.130、cudnn v7.4.1
• keras2.2.5

1.安装pycharm和Anaconda3-2019.10-Windows-x86_64

2.在pycharm软件的Terminal窗口操作；

3.创建虚拟环境

conda create -n DeepLearnFromTensorflow python=3.7
activate DeepLearnFromTensorflow  # win10下激活虚拟环境命令
conda env list

4.因为tensorflow库在国外服务器上，直接安装tensorflow会比较慢，则用清华镜像去安装。

​ 在电脑找到Anaconda Prompt可执行程序，进入相关窗口，加入清华源。

conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
conda config --set show_channel_urls yes

5.继续在Terminal窗口操作操作

pip install tensorflow-gpu==1.14.0 -i https://pypi.tuna.tsinghua.edu.cn/simple # 可以改成任意版本

​ 安装后，在python中输入import tensorflow，报错缺少CUDA 10.0

ImportError: Could not find 'cudart64_100.dll'. TensorFlow requires that this DLL be installed in a directory that is named in your %PATH% environment variable. Do
wnload and install CUDA 10.0 from this URL: https://developer.nvidia.com/cuda-90-download-archive

6.在提示的网址下载CUDA 10.0，我最终的下载地址如下：

https://developer.nvidia.com/cuda-10.0-download-archive?target_os=Windows&target_arch=x86_64&target_version=10&target_type=exelocal

​ 下载的CUDA为：cuda_10.0.130_411.31_win10.exe

7.点击cuda_10.0.130_411.31_win10.exe安装cuda，安装完成后，测试下是否安装成功：

​ 在C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\extras目录下进入cmd命令窗口，依次输入下面两个命令：

bandwidthTest.exe
deviceQuery.exe

​ 上面两个命令显示出Result = PASS信息，就表明cuda安装成功了。当然还可以用下面命令查看cuda版本。

C:\Users\msi-pc>nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2018 NVIDIA Corporation
Built on Sat_Aug_25_21:08:04_Central_Daylight_Time_2018
Cuda compilation tools, release 10.0, V10.0.130

8.tensorflow版本与CUDA，cuDNN版本一定要匹配，最好先查下自己的tensorflow适配的版本，在路径C:\Technology\Anaconda3-2019.10-Windows-x86_64\envs\DeepLearnFromTensorflow\Lib\site-packages\tensorflow\python\platform下看到build_info.py文件，例如我的tensorflow版本是1.14.0，其build_info.py文件的具体内容为：

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

is_cuda_build = True

cuda_version_number = '10.0'
cudnn_version_number = '7'
msvcp_dll_name = 'msvcp140.dll'
nvcuda_dll_name = 'nvcuda.dll'
cudart_dll_name = 'cudart64_100.dll'
cudnn_dll_name = 'cudnn64_7.dll'

​ 最好把pycharm环境重启下，否则版本匹配正确，也将cuda相关路径加入到环境变量中，在python中输入import tensorflow，依然报错缺少CUDA 10.0的问题。

9.在python中再输入import tensorflow，报错cudnn错误。

ImportError: Could not find 'cudnn64_7.dll'. TensorFlow requires that this DLL be installed in a directory that is named in your %PATH% environment variable. Note that installing cuDNN is a separate step from installing CUDA, and this DLL is often found in a different directory from the CUDA DLLs. You may install the necessary DLL by downloading cuDNN 7 from this URL: https://developer.nvidia.com/cudnn

​ 错误提示中，有下载cuDNN的链接，但是现在要下载cuDNN，点击下载的页面后都是出现要求先加入NVIDIA developers会员才能进行下载，但这个注册的过程非常慢，常常卡在第二个步骤。其具体下载cudnn的方法参考博客：https://blog.csdn.net/weixin_40392957/article/details/80207366

​ 1）我打开cuDNN各个版本的下载网址：https://developer.nvidia.com/rdp/cudnn-archive#a-collapse51b

​ 2）选择Download cuDNN v7.4.1 (Nov 8, 2018), for CUDA 10.0下的cuDNN Library for Windows 10，复制其链接地址：https://developer.nvidia.com/compute/machine-learning/cudnn/secure/v7.4.1.5/prod/10.0_20181108/cudnn-10.0-windows10-x64-v7.4.1.5.zip

​ 3）我电脑上没有迅雷，而我手机上有，我用手机下载好zip包后上传到电脑上，其cudnn-10.0-windows10-x64-v7.4.1.5.zip大小为208 MB

​ 4）解压cudnn-10.0-windows10-x64-v7.4.1.5.zip，其内容如下：

bin
include
lib
NVIDIA_SLA_cuDNN_Support.txt

​ 5）将cudnn的bin,include,lib中的内容 拷贝到C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0的bin,include,lib对应目录下，提示替换的话点击OK。完成，最好重启一下电脑。（参考博客：https://blog.csdn.net/c20081052/article/details/86683446）

10.测试安装的tensorflow-gpu环境是否成功，可运行下面程序：

​ 1）测试程序1

import tensorflow as tf
print(tf.__version__)
#查询tensorflow安装路径为:
print(tf.__path__ )

import os
from tensorflow.python.client import device_lib
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "99"

if __name__ == "__main__":
    print(device_lib.list_local_devices())

​ 程序输出：

C:\Technology\Anaconda3-2019.10-Windows-x86_64\envs\DeepLearnFromTensorflow\lib\site-packages\tensorflow\python\framework\dtypes.py:516: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint8 = np.dtype([("qint8", np.int8, 1)])
C:\Technology\Anaconda3-2019.10-Windows-x86_64\envs\DeepLearnFromTensorflow\lib\site-packages\tensorflow\python\framework\dtypes.py:517: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_quint8 = np.dtype([("quint8", np.uint8, 1)])
C:\Technology\Anaconda3-2019.10-Windows-x86_64\envs\DeepLearnFromTensorflow\lib\site-packages\tensorflow\python\framework\dtypes.py:518: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint16 = np.dtype([("qint16", np.int16, 1)])
C:\Technology\Anaconda3-2019.10-Windows-x86_64\envs\DeepLearnFromTensorflow\lib\site-packages\tensorflow\python\framework\dtypes.py:519: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_quint16 = np.dtype([("quint16", np.uint16, 1)])
C:\Technology\Anaconda3-2019.10-Windows-x86_64\envs\DeepLearnFromTensorflow\lib\site-packages\tensorflow\python\framework\dtypes.py:520: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint32 = np.dtype([("qint32", np.int32, 1)])
C:\Technology\Anaconda3-2019.10-Windows-x86_64\envs\DeepLearnFromTensorflow\lib\site-packages\tensorflow\python\framework\dtypes.py:525: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  np_resource = np.dtype([("resource", np.ubyte, 1)])
C:\Technology\Anaconda3-2019.10-Windows-x86_64\envs\DeepLearnFromTensorflow\lib\site-packages\tensorboard\compat\tensorflow_stub\dtypes.py:541: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint8 = np.dtype([("qint8", np.int8, 1)])
C:\Technology\Anaconda3-2019.10-Windows-x86_64\envs\DeepLearnFromTensorflow\lib\site-packages\tensorboard\compat\tensorflow_stub\dtypes.py:542: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_quint8 = np.dtype([("quint8", np.uint8, 1)])
C:\Technology\Anaconda3-2019.10-Windows-x86_64\envs\DeepLearnFromTensorflow\lib\site-packages\tensorboard\compat\tensorflow_stub\dtypes.py:543: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint16 = np.dtype([("qint16", np.int16, 1)])
C:\Technology\Anaconda3-2019.10-Windows-x86_64\envs\DeepLearnFromTensorflow\lib\site-packages\tensorboard\compat\tensorflow_stub\dtypes.py:544: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_quint16 = np.dtype([("quint16", np.uint16, 1)])
C:\Technology\Anaconda3-2019.10-Windows-x86_64\envs\DeepLearnFromTensorflow\lib\site-packages\tensorboard\compat\tensorflow_stub\dtypes.py:545: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint32 = np.dtype([("qint32", np.int32, 1)])
C:\Technology\Anaconda3-2019.10-Windows-x86_64\envs\DeepLearnFromTensorflow\lib\site-packages\tensorboard\compat\tensorflow_stub\dtypes.py:550: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  np_resource = np.dtype([("resource", np.ubyte, 1)])
1.14.0
['C:\\Technology\\Anaconda3-2019.10-Windows-x86_64\\envs\\DeepLearnFromTensorflow\\lib\\site-packages\\tensorflow\\python\\keras\\api\\_v1', 'C:\\Technology\\Anaconda3-2019.10-Windows-x86_64\\envs\\DeepLearnFromTensorflow\\lib\\site-packages\\tensorflow_estimator\\python\\estimator\\api\\_v1', 'C:\\Technology\\Anaconda3-2019.10-Windows-x86_64\\envs\\DeepLearnFromTensorflow\\lib\\site-packages\\tensorflow', 'C:\\Technology\\Anaconda3-2019.10-Windows-x86_64\\envs\\DeepLearnFromTensorflow\\lib\\site-packages\\tensorflow\\_api\\v1']
[name: "/device:CPU:0"
device_type: "CPU"
memory_limit: 268435456
locality {
}
incarnation: 2210577014675909570
, name: "/device:GPU:0"
device_type: "GPU"
memory_limit: 4997316608
locality {
  bus_id: 1
  links {
  }
}
incarnation: 8922325653411056018
physical_device_desc: "device: 0, name: GeForce GTX 1660 Ti, pci bus id: 0000:01:00.0, compute capability: 7.5"
]

解决FutureWarning: Passing (type, 1) or ‘1type’ as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type’的警告方法：

• （1）使用warning模块，但这会暂时忽略所有的FutureWarning报错。

  import warnings
  warnings.filterwarnings('ignore', category=FutureWarning)
  
  import tensorflow as tf
  

• （2）降低numpy版本，然后不在报上面的FutureWarning

  python
  如果Numpy的版本超过了1.17，那么就会出现这个警告;
  既然高版本（Numpy 1.17.0rc1)会报错，那么稍微降低版本不就可以了
  操作：
  pip uninstall numpy
  pip install numpy==1.16.4

  2）测试程序2

import tensorflow as tf
hello = tf.constant('Hello, TensorFlow!')
sess = tf.Session()
print(sess.run(hello))

程序输出：

WARNING:tensorflow:From D:/2021/DeepLearn/2.py:3: The name tf.Session is deprecated. Please use tf.compat.v1.Session instead.

2021-05-15 23:55:42.141233: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library nvcuda.dll
2021-05-15 23:55:42.164878: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1640] Found device 0 with properties: 
name: GeForce GTX 1660 Ti major: 7 minor: 5 memoryClockRate(GHz): 1.59
pciBusID: 0000:01:00.0
2021-05-15 23:55:42.165014: I tensorflow/stream_executor/platform/default/dlopen_checker_stub.cc:25] GPU libraries are statically linked, skip dlopen check.
2021-05-15 23:55:42.165140: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1763] Adding visible gpu devices: 0
2021-05-15 23:55:42.165414: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2021-05-15 23:55:42.167656: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1640] Found device 0 with properties: 
name: GeForce GTX 1660 Ti major: 7 minor: 5 memoryClockRate(GHz): 1.59
pciBusID: 0000:01:00.0
2021-05-15 23:55:42.167777: I tensorflow/stream_executor/platform/default/dlopen_checker_stub.cc:25] GPU libraries are statically linked, skip dlopen check.
2021-05-15 23:55:42.167884: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1763] Adding visible gpu devices: 0
2021-05-15 23:55:42.524395: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1181] Device interconnect StreamExecutor with strength 1 edge matrix:
2021-05-15 23:55:42.524489: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1187]      0 
2021-05-15 23:55:42.524540: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1200] 0:   N 
2021-05-15 23:55:42.524789: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1326] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 4765 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1660 Ti, pci bus id: 0000:01:00.0, compute capability: 7.5)
b'Hello, TensorFlow!'

11.在Terminal窗口的虚拟环境DeepLearnFromTensorflow下，安装keras，查看keras与tensorflow对应的版本，参考博客：https://blog.csdn.net/weixin_40109345/article/details/106730050

pip install keras==2.2.5

​ 安装好keras后，numpy库自动更新为1.20.3版本，然后我又重装为了1.16.4版本，但是报出如下错误：

scipy 1.6.3 requires numpy<1.23.0,>=1.16.5, but you have numpy 1.16.4 which is incompatible.

​ 将numpy库卸载后，安装numpy 1.16.5版本即可。

​