(from https://www.zhiyuanyao.com/deeplearning/2017/09/25/aws/\)

인스턴스 구동후 설치할 패키지

CarND-Vehicle-Detection-2에서 에러 발생

pip install opencv-python
conda install scikit-image

Launch Instance

Go to EC2 Dashboard
Click ”Launch Instance”
Click “Community AMI”
Search for AMI “udacity-dl”
Select AMI “udacity-dl”

Instance Type

Filter by “GPU instances”
Choose g2.2xlarge type
Click “Review and Launch”

Set Storage

Since some dataset can reach several GB, we need enough space to store it. Make sure your instance has 32GB storage

Click “Edit Storage”
Increase storage to 32GB
Click “Review and Launch” again

Security Group

In the course of Udacity-DLND, we typically run several jupyter notebooks, whose port by default is 8888. Thus, in order to visit this port, we need to modify the security group on AWS.

Click “Edit Security Group”
Set Security Group
    Choose “Create a new security group”
    Name the security group as “Jupyter”
    Click “Add Rule”
    Set “Type” to “Custom TCP Rule”
    Set “Port Range” to 8888
    Set “Source” to “Anywhere”
Click “Review and Launch” again

Once the security group “Jupyter” has been created, there is no need to create it again. From now on, whenever launching a new instance, just choose the security group “Jupyter”.

Launch

Click “Launch” and we are good to go with our GPU instance
    Right now, AWS will ask us whether we are using key to identify. Just go with “without key” choice since AMI has the

Login

Open terminal
Input: ssh [email protected], where X.X.X.X is the IPv4 Public IP we can find on AWS
Identification verification with password: “udacity”
If you want to change your password, use command passwd, and then you can use the new password to login the “udacity” instance

Run Your Code

Now that the instance has been setup, we can start running our notebook.

Clone our code to the server on AWS with, e.g. git clone https://github.com/ZhiyuanYaoJ/Udacity-DLND
Direct to one project with, e.g. cd Udacity-DLND/Projects/1-first-nn
Activate our environment source activate dl
Run the notebook jupyter notebook
Open the browser and visit X.X.X.X:8888, where X.X.X.X is the IP address of our EC2 instance
Now we are good to go!