The Trident is a collection of three super-module generators for Neural Network design and construction for biological sequences and beyond. The short paper explaining it is in progress.
This package contains 3 modules:
- prongs
- This contains the super-module generators called Prongs.
- preass
- This contains pre-assembled neural networks ranging from simple dense NNs to convolutional autoencoders
- utils
- This contains utilities used in the aforementioned modules.
- It also contains:
- A special Trainer that implements useful fitting methods such as batch flipping (for biological sequences) and Epoch Sampling (Binning Methods paper)
- A Predictor for producing predictions
- A candidate scorer and candidate repeater which are used for hyperparameter optimization with poseigen_compass
- Binned loss which is using bin metrics as a loss function (Binning Methods paper)
- Synthetic data generation (see DevLoss case study)