MouffetΒΆ
Mouffet is a framework the provides a strong and flexible structure for training and evaluating models for reproducible science.
A common problem with developping models is that it is generally an iterative problem. To achieve the desired results, several models are usually created and evaluated, with different settings each. Keeping track of which model used which settings can quickly become a problem. If several methods of evaluation are also used, this can soon become problematic.
With the multiplication of model creation tools, each with their own interface, it has also become more and more complicated to compare different models.
Finally, often in science only the final results are reported, and not always in a proper manner. This can lead to results that cannot be easily reproduced and thus to bad science.
With mouffet, the goal is to define some core concepts surrounding model creation and evaluation and provide an interface to these concepts. Reference classes are provided and should be subclassed to ensure a unified entry into model creation and evaluation.
While initially developed to work with deep learning models, the concepts surrounding the package are universal and can be used with any type of models. Mouffet was also designed to be platform and model framework agnostic. This means that several models created with different frameworks (e.g. tensorflow, pytorch, etc.) could be trained and compared when encapsulated with mouffet.
Core concepts
API Documentation
Configuration options
Examples