Cheesecake is a visual programming environment for AI model building, written in C#.
- Customisable model architectures - layer units, weight and bias initialisations, activations, etc
- Choose your dataset! It currently must be supplied in a specified format (as detailed in
Data.cs), but work is ongoing to allow custom delimiters in text files - Model training with different batch sizes, cost functions, learning rates, etc
- Support for multi-branch layer topologies
- Parameter loading and saving
- Written with minimal/no use of external modules
- GUI - I will be using WPF for my GUI after I finish implementing the backend (not much more to go!).
- Metrics - it's fairly simple to implement basic metrics such as accuracy.
- More layer types. This is my primary long-term ambition - I'm intending that Cheesecake becomes a tool for beginners to sharpen their intuition surrounding model building, which is so far only possible through the fully connected layers I have implemented. I'd like to implement recurrent (RNNs, LSTMs, GRUs) and CNN layers in the future to broaden use cases.
- Support for more dataset types. This links strongly to the previous aim - more layer types implies support for a wider range of datasets, such as audio or images.
It's pretty sweet!