Welcome to hidten’s documentation!

hidten

A framework that integrates hidden Markov and related models with modern deep learning—revealing hid(den) structure within your ten(sors).

Find the documentation here to get started.

• Provides intuitive classes to easily define latent state graphs and combine emission distributions for both discrete and continuous input tracks. 🧩

• Implements core Hidden Markov Model (HMM) algorithms for training and inference, including Forward–Backward and Viterbi. ⚡

• Enables gradient-based, highly parallel training as well as efficient parallel inference. 🔄

• Pythonic, clean, extensible, and thoroughly tested. ✅

Installation

python -m pip install hidten[tensorflow]

Installation for developers

git clone https://github.com/Gaius-Augustus/hidten pip install -e .[tensorflow,plots,test,doc]

Build the docs:

cd docs && make html

License

This project is licensed under the MIT license.

Tools using hidten

• Tiberius: End-to-End Deep Learning with an HMM for Gene Prediction

• learnMSA: deep protein multiple alignments with large language and hidden Markov models

• bricks2marble: Pre- and postprocessing tools for genome annotation

User Guide:

• Getting Started
• Algorithms for Inference and Training
  • Emission scores
  • Likelihood
  • Posterior
  • Viterbi
  • Maximum expected accuracy
• Emitters
  • Categorical
  • Continuous
  • Padding
  • Allowing and sharing
  • Creating Custom Emitters
• Transitioners
  • Built-in Transitioner
  • Sharing
  • Creating Custom Transitioners
• Plotting
  • Quick start
  • Fine-tuned plotting
  • Supported emitter types
  • Example notebook
  • API reference
• Priors
• Advanced
  • Parallel algorithms

API Reference:

• API Reference
  • Core Module
  • HMM Classes
  • Configuration
  • Emitters
  • Transitioners
  • Priors
  • Generic Base Classes
  • Utilities
• TensorFlow Backend
  • Core TensorFlow HMM
  • TensorFlow Emitters
  • TensorFlow Transitioners
  • TensorFlow Priors
  • Algorithms
  • TensorFlow Utilities