Build AI / machine learning environment with Python

■ PC environment CPU：i7 Main memory: 16GB GPU：NVIDIA 「Tesla80」

■OS WindowsServer2019／64bit

■ Setup tool ● GPU environment (Cuda) Driver "Tesla K80" CUDA toolkit CUDA10.0 [cuda_10.0.130_win10_network.exe] cuDNN7.4 [cudnn-10.0-windows10-x64-v7.6.0.64.zip] ● Python environment Python3.8 (Miniconda3 Windows 64-bit)

① GPU (Cuda) driver setup Download and install the Tesla K80 driver from NVIDIA.

URL：https://www.nvidia.com/Download/Find.aspx Select the following product type from the site Product Type: Datacenter/Tesla Product Series: K-Series Product Type: Tesla k80 OS: Windows 2019

② Cuda Toolkit setup ● CUDA Toolkit download site Download CUDA10.0 [cuda_10.0.130_win10_network.exe] from the following site https://developer.nvidia.com/cuda-10.0-download-archive

Install the downloaded exe Select "A GREE AND CONTINUE" on the screen below

Select Custom and click Next

Check below and click Next

Click Next

Check and click Next

Click Next

Click CLOSE

● cudnn download site https://developer.nvidia.com/rdp/cudnn-archive Download cuDNN7.4 [cudnn-10.0-windows10-x64-v7.6.0.64.zip] from the following site https://developer.nvidia.com/compute/machine-learning/cudnn/secure/v7.6.0.64/prod/10.0_20190516/cudnn-10.0-windows10-x64-v7.6.0.64.zip

Unzip the DL "cudnn-10.0-windows10-x64-v7.6.0.64.zip" Copy the file according to the configuration below

● Setup cudnn-10.0-windows10-x64-v7.6.0.64 └─cuda │ ├─bin │ cudnn64_7.dll => Copy to C: \ Program Files \ NVIDIA GPU Computing Toolkit \ CUDA \ v10.0 \ bin
│ ├─include │ cudnn.h => Copy to C: \ Program Files \ NVIDIA GPU Computing Toolkit \ CUDA \ v10.0 \ include │ └─lib └─x64 copy to cudnn.lib => C: \ Program Files \ NVIDIA GPU Computing Toolkit \ CUDA \ v10.0 \ lib \ x64

This completes the CUDA Toolkit setup

③ Python (Anaconda or miniconda) environment This time, we will set up a mini version of Anaconda, Miniconda, and build a Python virtual environment. Download "Miniconda3 Windows 64-bit (Miniconda3-latest-Windows-x86_64.exe)" from the following site https://conda.io/miniconda.html

Double-click the downloaded "Miniconda3-latest-Windows-x86_64.exe" and follow the steps below to install it. In addition, since the user name is "Administrator", Path is replaced with your own PC environment.

● Press Next

● Select I Agree

● Press Next

● Press Next

● Press Install There is a check box as to whether to change the global path, but this is not changed as recommended (do not check)

● Press Next

● Uncheck the check box and press Finish

● Start Anaconda Prompt (miniconda3) from the start menu

Anaconda Prompt is displayed

Subsequent work will be done with Anaconda Prompt

■ Create any version of Python environment As of October 2020, the latest version of Python is 3.8.5, but the latest version of Tensorflow does not support 3.8, so it was created with the "3.7" version.

● Execution command

conda create -n py37 python=3.7

● A list of additional installation packages will be displayed. Enter "y" and press enter.

When the installation is completed, the virtual environment will be created in the following path by default. 「C:\ProgramData\Miniconda3\envs\py37」

● Start virtual environment

conda activate py37

● python version check

python --version

● Check the created virtual environment

conda info -e

● When stopping the virtual environment

conda deactivate

■ Installation of extension packages python -m pip install --upgrade pip pip install tensorflow-gpu==2.0 pip install keras==2.3.1 pip install scikit-learn pip install pandas pip install beautifulsoup4 pip install requests pip install matplotlib pip install pillow pip install opencv-python pip install pyttsx3 pip install pydot pip install pydotplus pip install lxml pip install janome pip install seaborn pip install pydot-ng pip install IPython pip install jupyter conda install graphviz

Put the path of the library "graphviz" installed in the virtual environment in the system environment variable "Path" Example) C: \ ProgramData \ Miniconda3 \ envs \ py37 \ Library \ bin \ graphviz

■ Start Jupyter Notebook

jupyter notebook

The browser starts and the following screen is displayed

■ Set the source code to be executed below C:\share

■ Change the initial directory of Jupyter Notebook

1. Create a jupyter notebook configuration file Launch virtual environment with Anaconda Prompt

conda activate py37

Execute the following command

jupyter notebook --generate-config

Since "default config" is generated in [C: \ Users \ [user name] \ .jupyter \ jupyter_notebook_config.py], "default config" c.NotebookApp.notebook_dir = '' Is modified to the following ("c: \ share" this time) * Delete the # at the beginning c.NotebookApp.notebook_dir = 'c:\share'

This completes the setting

■Jupyter Notebook　tips

・ Help ESC＋H

-Reset the program execution of Jupyter Notebook Initialize the kernel with "Kernel → Restart & Clear Output" on the toolbar

-Display / cancel cell line number ESC+L