poseigen-trident 0.1.5

Neural network design and training utilities for the Poseigen family

Trident

Trident is the neural network design and training package in the Poseigen family. It provides super-module generators, pre-assembled architectures, and a full training toolkit for biological sequences and beyond.

Features

Trident is organized into five modules:

• prongs: super-module generators (Prongs) for composable neural network construction.
• preass: pre-assembled neural networks ranging from simple dense networks to convolutional autoencoders.
• deepstarr: DeepSTARR-based model implementations.
• othermodels: additional model architectures.
• utils: shared training and evaluation utilities including:
  • A Trainer with batch flipping and Epoch Sampling (Binning Methods paper).
  • A Predictor for producing predictions.
  • Candidate scorer and repeater for hyperparameter optimization with poseigen_compass.
  • Binned loss using bin metrics as a loss function.
  • Synthetic data generation.

Installation

Install from PyPI:

pip install poseigen_trident

For local development, install from source using your preferred editable-install workflow.

Usage

Import modules directly:

import poseigen_trident.utils as tu
import poseigen_trident.prongs as prongs
import poseigen_trident.preass as preass

Project Status

poseigen_trident is in active development and is intended to support neural network workflows across the Poseigen ecosystem.

Related Projects

• poseigen_seaside: shared utilities and metrics foundation.
• poseigen_binmeths: binning and split-generation utilities.
• poseigen_compass: hyperparameter optimization and model evaluation.

License

This project is released under the MIT License.