ttglow 0.1.3

Tensor Train linear algebra package for efficient high-dimensional tensor operations

TTGlow

Tensor Train linear algebra built on PyTorch for efficient high-dimensional tensor operations.

Overview

TTGlow provides an efficient implementation of the Tensor Train (TT) decomposition, a powerful technique for representing and manipulating high-dimensional tensors with reduced memory footprint and computational cost. This package is particularly useful for:

• High-dimensional data analysis
• Quantum many-body physics simulations
• Machine learning with tensor methods
• Efficient storage and computation of large-scale tensors

Installation

Using pip (Recommended)

pip install ttglow

From source (for development)

# Clone the repository
git clone https://github.com/romanellerbrock/ttglow.git
cd ttglow

# Install with pixi
pixi install
pixi shell

# Or install with pip
pip install -e .

Quick Start

import torch
from ttglow import TensorTrain

# Convert a dense tensor to TT format
dense = torch.randn(8, 8, 8, 8, dtype=torch.float64)
tt = TensorTrain.from_dense(dense, max_rank=4)  # Compress with rank truncation

print(f"Original: {dense.numel():,} elements")
print(f"Compressed: {tt.ttnumel:,} elements ({tt.ttnumel/dense.numel():.1%})")

# Arithmetic with operator overloading
tt2 = TensorTrain.random([8, 8, 8, 8], ranks=[4, 4, 4])
result = tt + tt2       # Addition
result = tt * tt2       # Hadamard (elementwise) product
result = 2.5 * tt       # Scalar multiplication
inner = tt @ tt2        # Inner product

# Shape operations (PyTorch-like API)
tt_reshaped = tt.reshape([64, 64])
tt_transposed = tt.transpose(0, 2)
tt_flat = tt.flatten(1, 3)

# Convert back to dense
reconstructed = tt.to_tensor()

Features

• Efficient TT representation: Store high-dimensional tensors with minimal memory
• Core operations: dot, hadamard, add, scale
• Shape operations: reshape, transpose, permute, squeeze, unsqueeze, flatten, cat, stack
• Decomposition: from_dense (TT-SVD with optional rank truncation), svd, qr
• TT-Matrix support: Matrix-vector products, operator composition, kron, kronadd
• PyTorch-like API: .to(), .clone(), .detach(), familiar tensor interface
• PyTorch backend: Full GPU support and automatic differentiation

Development

Running Tests

pixi run test

Code Formatting

pixi run format

Linting

pixi run lint

Test Coverage

pixi run test-cov

Project Structure

ttglow/
├── src/ttglow/           # Main package code
│   ├── __init__.py
│   ├── tensortrain.py    # TensorTrain class and operations
│   ├── ttmatrix.py       # TT-Matrix (MPO) support
│   └── linalg.py         # Linear algebra (SVD, QR, norms)
├── tests/                # Test suite
├── examples/             # Usage examples
├── docs/                 # Documentation website
└── pyproject.toml        # Package configuration

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

References

• Oseledets, I. V. (2011). "Tensor-train decomposition". SIAM J. Sci. Comput., 33(5), 2295-2317.