SEADS Machine Manual

Running Tensorflow

1. Run "$ ssh user_id@ip_address"
2. Enter password
3. Run Tensorflow

• (without Jupyter, directly to shell prompt)

  1. Run "$ nvidia-docker run -it -p port1:8888 -p port2:6006 --name name_00 -v path:/data tf:seads /bin/bash"
  • Use the allocated port numbers for "port1" and "port2"
  • Use a distinguishable name for your created container instead of "name_00" (For each run, the name should be different)
  • "path" is the path (e.g. "/home/user_id") you want to connect to the created container, which will be located under "/data" of the container
  • Tensorboard will be available by connecting "http://ip_address:port2" in your web browser.
  1. You are now in the created container where you can use the shell just as the python virtual environment for Tensorflow.

• (with Jupyter)

  1. Run "$ nvidia-docker run -it -p port1:8888 -p port2:6006 --name name_00 -v path:/notebooks tf:seads"
  2. Copy the characters following "/?token="
  3. Connect "http://ip_address:port1" in your web browser and enter the copied token
  4. Now you see Jupyter interface

Running OpenCV

1. Run "$ ssh user_id@ip_address"
2. Enter password
3. Create a container with OpenCV

• (without Jupyter, directly to shell prompt)

  1. Run "$ docker run -it --name name_00 -v path:/data floydhub/dl-opencv:latest-gpu-py2 /bin/bash"
     * Use a distinguishable name for your created container instead of "name_00" (For each run, the name should be different)
     * "path" is the path (e.g. "/home/user_id") you want to connect to the created container, which will be located under "/data" of the container

• (with Jupyter)

  1. Run "$ docker run -it --name name_00 -v path:/notebooks floydhub/dl-opencv:latest-gpu-py2"
  2. Copy the characters following "/?token="
  3. Connect "http://ip_address:port1" in your web browser and enter the copied token
  4. Now you see Jupyter interface

Closing

• Push (Cntrl + c) twice in the terminal
• If you did not finish your container properly, type "$ docker container stop name_00" where "name_00" is the container's name you made when creating the container.
  • When you cannot remember the name correctly, you can list all running containers by using "$ docker container ls"

Tips

• When you like to keep the server computer to run even after disconnected from your local computer

  1. Run "$ screen" just after the second step for < Running > and continue next steps
  2. Push (Ctrl + a, d) at any time when you want to disconnect from the ssh
  3. Run "$ screen -r" just after the second step for Running, when you want to return to the work

• Large HDD storage is available in "/data"

Questions?

Please send an email to [email protected]