Remokaggle 🚀

A quick script to setup SSH to Kaggle Kernel for Deep Learning. In order to use that sexy P100 for free (and without many restriction of Jupyter Notebook) :P

‼️‼️ UPDATE ‼️‼️

Kaggle had blocked any attempt of creating an SSH tunneling into a running Kaggle kernel. So unfortunately this repo is NO LONGER WORKS. Attempting to use the script presented in this repo might violate Term and Conditions of Kaggle and WILL GET YOUR KAGGLE ACCOUNT BANNED!

While we wait for Kaggle to change its policy, the best solution for a free (or relatively cheap) GPU instance with a full SSH connection is GOOGLE COLAB (As well as Colab Pro) with a Tesla V100 for up to 24 hours. I have created another package that can help you get an SSH tunnel to a running Colab notebook in this repo: https://github.com/lamhoangtung/colab_ssh. You guys should use it instead.

Thanks for checking by ✌️ !

Preparation

1. Go to https://dashboard.ngrok.com/auth and get your authentication token after register with Google or Github (You only have to do this once)
2. Modify the get_ssh.py file:
   1. Replace your ngrok authentication token at line 7 in
   2. Set a password for your user at line 10 (just in case you need it, we will use SSH Key authentication from now on)
   3. Create a public Github Gist with file name authorized_keys and paste your public SSH key (usually located at ~/.ssh/id_rsa.pub) as the content of the gist.
   4. If you don't have an SSH key (RSA Key Pair), please refer to the first step in this article
   5. Replace the link to raw content of your gist at line 11

How to setup SSH connection

• Step 1: Create a new Kaggle kernel

• Step 2: Go to Kernel Settings and turn on GPU and Internet

  

• Step 3: Copy the whole content of the modified get_ssh.py and paste to the kernel as the first block of code.

  

• Step 4: Click Commit on the top right conner and wait a minutes for everything to set up. You should see a pop-up windows like this:

  

  Save the host address information tcp://0.tcp.ngrok.io:16360 to SSH. If some how you forget the host address and port, go to your Ngrok Dashboard -> Status and you will found them again.

How to SSH

• ssh root@0.tcp.ngrok.io -p 16360 <-- port number dictated in above output
• With root password also dictated in above output
• You won't have to use the password since we already used SSH key authentication to logging into your server
• You can turn off the browser, disconnect from the server and it will still running. But keep in mind this server can only live for 9 hours at max.

How to quickly setup the server

• Kaggle Kernel already provide many Machine Learning and Deep Learning package and library ready to use with a powerful NVIDIA Tesla P100 GPU so you might be good to go. But if you want something more that I usually use. Run this to get more:

  cd ~
  wget https://github.com/lamhoangtung/kaggle-kernel-setup/raw/master/install_common.sh
  chmod 777 install_common.sh
  ./install_common.sh

  And follow the instruction.

• You only need to Commit again in the same kernel to get everything up and running again after terminate the server. Host address information will be reset each time you commit.

• You will need multiple Kaggle account and multiple ngrok account to get multiple server running at the same time.

How to terminate the server

• Go to https://www.kaggle.com/kernels and look at the Your Recent Kernels section and hit Stop.

  

• If you encounter any problem while setting up the connection. Please terminate the server as above. Then open the kernel in editor mode and go to Run > Power off, then turn it on manually.

  

• Keep in mind everything will be lost (including process and files, ...) when you Power off or Terminate the kernel

Feels free to contribute to this tiny repo since I don't have much experience with bash and linux ;). All Pull Request are welcomed ❤️