ML-Boilerplate

Boilerplate repository for machine learning projects. This assumes a primarily scripting-based setup, in which most interaction is implemented using a central file src/run.py that loads/processes data and loads/trains/tests models defined in separate modules.

Directory structure

• vis/: all types of visualisations, like plots and images of model architectures
• model/: model related outputs, like training and testing logs, model weights and results
• src/: contains all source code in different subdirectories:
  • models/: code that implements model architectures
  • notebooks/: Jupyter Notebooks for quick testing and visualization that are not part of the main workflow
  • utils/: utility modules:
    • model/: modules for loading, training and testing your models
    • data/: modules for loading and preprocessing/augmenting your data
    • generic.py: module for generic helper functions, e.g. status printing for scripts
  • run.py: main file in which to set up the logical flow of the project and run different interactions, i.e. call the functions defined in the different modules above

Note: there is no central data/ directory as data might be placed outside of this repository. If not, I recommend to place it at the root.