/basic-deep-learning-tools

The repo was created as a part of a student round discussino at Fraunhofer AISEC

Primary LanguagePython

basic-deep-learning-tools

The repo was created as a part of a student round discussion at Fraunhofer-AISEC

Steps

  • Create a conda environment using
    conda create -n basic_torch python=3.10
  • In order to run this repo,you can directly create the clone of the same environment using the following command(From directory root):
    conda env create --file basic_torch.yml
  • If you install more dependencies or change version of already existing libraries, make sure to update the enrionment file using:
  conda env export --name basic_torch > basic_torch.yml

Tips to use on server

  • Run the training using the screen command so that in case your connection times out and you're disconnected, your training won't halt
  • Some commonly used screen commands
   # Name your screen in case you are running multiple python scripts
   screen -S my_screen_name
   
   # To disconnect from the screen without killing the process press **Ctrl + a, d**
   
   # To reconnect with the screen
   screen -r my_screen_name
   
   # To list all the running screen
   screen -ls
  • Setup Pycharm or VsCode with remote Python interpreter to check if your local changes work with the libraries installed in the server conda env
  • In case you don't use PyCharm or VSCode, then using pudb is a good option for debugging directly from the terminal in case your code breaks on the server
  • There are many ways to organise a deep learning repo but the basic structure I prefer is for a barebones repo is:
 REPO_ROOT
├── basic_torch.yml
├── config.yml
├── data
│ └── MNIST
├── model.py
├── README.md
├── train.py
└── utils.py
  • Conda autocomplete bashscript link