PyTau

Powerful Changepoint Detection for Neural Data

[](https://joss.theoj.org/papers/e3e3d9ce5b59166cef17ee7e9bb9f53c) [](https://results.pre-commit.ci/latest/github/abuzarmahmood/pytau/master) [](https://github.com/abuzarmahmood/pytau/actions/workflows/pytest_workflow.yml)

🚀 What is PyTau?

PyTau is a specialized Python package for detecting state changes in neural data using Bayesian changepoint models. It provides:

• Streamlined batch inference on PyMC3-based changepoint models
• Robust detection of state transitions in neural firing patterns
• Comprehensive visualization tools for model results

📚 Documentation

Full API Documentation

⚡ Quick Start

```bash

Create and activate conda environment

conda create -n "pytauenv" python=3.6.13 ipython notebook -y conda activate pytauenv

Clone repository

git clone https://github.com/abuzarmahmood/pytau.git

Install package

cd pytau pip install -e . # Install in development mode

Alternative: Install from requirements file

pip install -r requirements.txt

Download test data and run example notebook

cd pytau/howto bash scripts/downloadtest_data.sh cd notebooks jupyter notebook ```

🧠 Key Features

• Multiple Model Types:

  • Single-taste Poisson models
  • All-taste hierarchical models
  • Dirichlet process models for automatic state detection
  • Variable sigmoid models for flexible transition shapes

• Flexible Data Handling:

  • Support for shuffled, simulated, and actual neural data
  • Comprehensive preprocessing pipeline
  • Automated model selection

• Powerful Analysis Tools:

  • State transition detection
  • Cross-region correlation analysis
  • Statistical significance testing

📊 Data Organization

PyTau uses a centralized database approach for model management:

• Centralized Storage: Models stored in a central location with metadata
• Comprehensive Tracking: Each model includes:
  • Animal and session information
  • Region and taste details
  • Model parameters and preprocessing steps
  • Fit statistics and results

🔄 Pipeline Architecture

PyTau's modular pipeline ensures reproducible analysis:

1️⃣ Model Generation

• Purpose: Creates and fits Bayesian changepoint models
• Components: Various model types for different neural data structures
• Input: Processed spike trains and model parameters
• Output: Fitted model with posterior samples

2️⃣ Data Preprocessing

• Purpose: Prepares raw neural data for modeling
• Features: Handles various data transformations (shuffling, simulation)
• Input: Raw spike trains with parameters for processing
• Output: Binned, processed spike count data

3️⃣ I/O Management

• Purpose: Handles data loading, model storage, and database management
• Features: Automatic model retrieval or fitting based on parameters
• Operations: HDF5 data loading, metadata tracking, database integration

4️⃣ Batch Processing

• Purpose: Enables large-scale model fitting across datasets
• Features: Parameter iteration and parallel processing

💻 Advanced Usage

Parallelization

PyTau supports parallel processing using GNU Parallel with isolated Theano compilation directories to prevent clashes. See single_process.sh and this implementation.

🤝 Contributing

We welcome contributions to PyTau! Please see the CONTRIBUTING.md file for guidelines on how to contribute to the project.

📜 Citation

If you use PyTau in your research, please cite: @article{pytau2023, title={PyTau: Bayesian Changepoint Detection for Neural Data}, author={Mahmood, Abuzar}, journal={Journal of Open Source Software}, year={2023} }