Yi-Min's Blog

Stay Hungry Stay Foolish

0%

Machine Learning Ubuntu 18.04 環境建置手札

Posted on Edited on In Disqus:
Symbols count in article: 2.8k Reading time ≈ 3 mins.

重新安裝ssh server

1
2
3
sudo apt-get remove openssh-server
sudo reboot
sudo apt-get install openssh-server

ssh允許使用root登入

1
2
sudo vim /etc/ssh/sshd_config
# change "PermitRootLogin no" to "PermitRootLogin yes"

ssh免密碼登入

建立ssh rsa key

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
[neil@yimincai.net ~]$ ssh-keygen -t rsa

Generating public/private rsa key pair.
Enter file in which to save the key (/home/neil/.ssh/id_rsa): [Press enter key]
Created directory '/home/neil/.ssh'.
Enter passphrase (empty for no passphrase): [Press enter key]
Enter same passphrase again: [Press enter key]
Your identification has been saved in /home/neil/.ssh/id_rsa.
Your public key has been saved in /home/neil/.ssh/id_rsa.pub.
The key fingerprint is:
1f:ad:49:00:4a:d5:ab:99:b3:b0:f9:09:91:c4:ed:d2 neil@yimincai.net
The key's randomart image is:
+--[ RSA 2048]----+
|        ..oooE.++|
|         o. o..  |
|          ..   . |
|         o  . . o|
|        S .  . + |
|       . .    . o|
|      . o o    ..|
|       + +       |
|        +.       |
+-----------------+

Create .ssh Directory on host server

1
2
3
4
5
6
7
[neil@yimincai.net ~]$ ssh alyssa@192.168.50.6 mkdir -p .ssh

The authenticity of host '192.168.50.6 (192.168.50.6)' can't be established.
RSA key fingerprint is 48:0e:28:11:d6:81:68:16:04:5f:db:49:02:la:12:4e.
Are you sure you want to continue connecting (yes/no)? yes
Warning: Permanently added '192.168.50.6' (ECDSA) to the list of known hosts.
alyssa@192.168.50.6's password: [Enter Your Password Here]

Copy public key to host server

1
cat ~/.ssh/id_rsa.pub | ssh alyssa@192.168.50.6 'cat >> .ssh/authorized_keys'

重新啟動ssh服務

1
service ssh restatus

確認ssh服務狀況

1
service ssh status

防火牆設定允許ssh服務

1
2
3
sudo ufw allow ssh
# or
sudo ufw allow 22

設定Python環境

1. 使用Python3

1
sudo apt-get install python3-venv

建立Python3虛擬環境

1
2
# 在當前資料夾建立環境
python3 -m venv <venvName>

啟動環境

1
source ./<venvName>/bin/activate

2. 使用Conda作為Python環境

(Option)移除Python 3.6

⚠ 不要輕易移除，曾試過移除後GUI失效

1
sudo apt-get remove python3.6

移除Python 3.4及其dependent

1
sudo apt-get remove --auto-remove python3.6

清除Python 3.4

1
2
3
sudo apt-get purge python3.4
# or
sudo apt-get purge --auto-remove python3.6

安裝 Conda for Python 3.7 64bit

可以到Conda Docs頁面確認下載版本或使用下方範例(Python 3.7)

1
2
cd ~/Download
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh

wget-miniconda

更改檔案權限

1
sudo chmod +x ./Miniconda3-latest-Linux-x86_64.sh

安裝

1
./Miniconda3-latest-Linux-x86_64.sh

install-miniconda

同意license terms

agree-license-terms

Hit Enter

conti-install-conda

Type “yes”

run-conda-init

重啟Terminal、測試command

1
conda info

conda-info

建立Conda venv（虛擬環境）

確認安裝了多少個虛擬環境

1
conda env list

假設建一個名為 ml 的 venv 並設定Python 為3.6

1
conda create --name ml python=3.6

Press y

create-venv-y

1
2
3
4
# To activate this environment, use
conda activate ml
# To deactivate an active environment, use
conda deactivate

刪除虛擬環境或package

1
conda remove --name ml numpy

刪除虛擬環境

1
conda env remove --name myenv

安裝 Machine Learning 相關環境

安裝常用環境，可參考下方文章

1
2
3
4
5
6
7
8
9
10
11
12
pip install numpy
pip install pandas
pip install scikit-learn
pip install --upgrade tensorflow
pip install Keras
pip install Pillow
pip install tensorflow
pip install torch
.
.
.
etc.

⚠ 若在 import tensorflow as tf 時出現： Illegal instruction (core dumped)經確認是CPU缺少avx指令集的關係（CPU架構太舊不支援）。

可降低Tensorflow版本解決

1
2
3
pip uninstall tensorflow
# and then
pip install tensorflow==1.4