放寒假了，又到了折騰新環境的時間。截止目前（2020年1月15日），TensorFlow的最新版本爲2.1.0，PyTorch的最新版本爲1.3.1。作爲強迫症，也爲可能的工程化考慮，希望能夠把它們安裝在同一個環境中。
基本環境：
- windows10
- Python3.6
首先在Anaconda中新建一個名叫cv的Conda環境：
conda create -n cv python=3.6
爲了加快安裝速度，將conda源換成清華源。.condarc中的內容如下：
channels:
- https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/pytorch/
- https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/
- https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
- defaults
show_channel_urls: true
要安裝PyTorch，注意先看一下可供安裝的版本：
conda search pytorch=1.3.1
Loading channels: done
# Name Version Build Channel
pytorch 1.3.1 cpu_py36h9f948e0_0 anaconda/pkgs/main
pytorch 1.3.1 cpu_py36h9f948e0_0 pkgs/main
pytorch 1.3.1 cpu_py37h9f948e0_0 anaconda/pkgs/main
pytorch 1.3.1 cpu_py37h9f948e0_0 pkgs/main
pytorch 1.3.1 py3.5_cpu_0 anaconda/cloud/pytorch
pytorch 1.3.1 py3.5_cuda101_cudnn7_0 anaconda/cloud/pytorch
pytorch 1.3.1 py3.5_cuda92_cudnn7_0 anaconda/cloud/pytorch
pytorch 1.3.1 py3.6_cpu_0 anaconda/cloud/pytorch
pytorch 1.3.1 py3.6_cuda101_cudnn7_0 anaconda/cloud/pytorch
pytorch 1.3.1 py3.6_cuda92_cudnn7_0 anaconda/cloud/pytorch
pytorch 1.3.1 py3.7_cpu_0 anaconda/cloud/pytorch
pytorch 1.3.1 py3.7_cuda101_cudnn7_0 anaconda/cloud/pytorch
pytorch 1.3.1 py3.7_cuda92_cudnn7_0 anaconda/cloud/pytorch
注意到PyTorch1.3.1對應的cuda版本是10.1或9.2，因此係統中的cuda版本不能低於這個。
安裝PyTorch1.3.1：
conda install pytorch=1.3.1=py3.6_cuda101_cudnn7_0 torchvision cudatoolkit=10.1
由於網絡原因，下載部分包時可能會出現HTTP Error，重複上述命令，直到安裝成功爲止。
接下來安裝TensorFlow。先看一下可供安裝的版本
conda search tensorflow=2
Loading channels: done
# Name Version Build Channel
tensorflow 2.0.0 eigen_py36h457aea3_0 anaconda/pkgs/main
tensorflow 2.0.0 eigen_py36h457aea3_0 pkgs/main
tensorflow 2.0.0 eigen_py37hbfc5123_0 anaconda/pkgs/main
tensorflow 2.0.0 eigen_py37hbfc5123_0 pkgs/main
tensorflow 2.0.0 gpu_py36hfdd5754_0 anaconda/pkgs/main
tensorflow 2.0.0 gpu_py36hfdd5754_0 pkgs/main
tensorflow 2.0.0 gpu_py37h57d29ca_0 anaconda/pkgs/main
tensorflow 2.0.0 gpu_py37h57d29ca_0 pkgs/main
tensorflow 2.0.0 mkl_py36h781710d_0 anaconda/pkgs/main
tensorflow 2.0.0 mkl_py36h781710d_0 pkgs/main
tensorflow 2.0.0 mkl_py37he1bbcac_0 anaconda/pkgs/main
tensorflow 2.0.0 mkl_py37he1bbcac_0 pkgs/main
可見conda源中的TensorFlow只更新到2.0.0，因此用pip安裝，默認安裝最新版本：
pip install tensorflow
安裝完成後import tensorflow as tf時遇到了問題：
ImportError: DLL load failed: The specified module could not be found.
在Github上找到了解決方法：https://github.com/tensorflow/tensorflow/issues/35749
方法一：TensorFlow降到2.0.0 【O__O "…】
方法二：下載並安裝：https://support.microsoft.com/en-us/help/2977003/the-latest-supported-visual-c-downloads
親測方法二可行
最後測試一下是否能用cuda：
import torch
print(torch.cuda.is_available())
print(torch.version.cuda)
import tensorflow as tf
print(tf.test.is_gpu_available())
輸出：
True
10.1
2020-01-15 20:52:48.968059: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
WARNING:tensorflow:From E:/0WorkSpace/PyCharm_Projects/Kaggle/Bengali/test.py:11: is_gpu_available (from tensorflow.python.framework.test_util) is deprecated and will be removed in a future version.
Instructions for updating:
Use `tf.config.list_physical_devices('GPU')` instead.
2020-01-15 20:52:51.762533: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2020-01-15 20:52:51.763974: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library nvcuda.dll
2020-01-15 20:52:51.765366: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce RTX 2060 computeCapability: 7.5
coreClock: 1.2GHz coreCount: 30 deviceMemorySize: 6.00GiB deviceMemoryBandwidth: 312.97GiB/s
2020-01-15 20:52:51.765712: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2020-01-15 20:52:51.765887: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2020-01-15 20:52:51.766218: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cufft64_10.dll
2020-01-15 20:52:51.766447: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library curand64_10.dll
2020-01-15 20:52:51.804424: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusolver64_10.dll
2020-01-15 20:52:51.804613: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusparse64_10.dll
2020-01-15 20:52:51.804787: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2020-01-15 20:52:51.805795: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1697] Adding visible gpu devices: 0
2020-01-15 20:52:53.983713: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1096] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-01-15 20:52:53.983903: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102] 0
2020-01-15 20:52:53.984016: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] 0: N
2020-01-15 20:52:53.986073: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1241] Created TensorFlow device (/device:GPU:0 with 4604 MB memory) -> physical GPU (device: 0, name: GeForce RTX 2060, pci bus id: 0000:01:00.0, compute capability: 7.5)
True
完結撒花花~~~
