torch-dxdt Documentation

torch-dxdt - PyTorch Differentiable Numerical Differentiation

A PyTorch implementation of numerical differentiation methods for noisy time series data.

Contents

Features

  • πŸ”₯ Fully Differentiable: All methods support PyTorch autograd for backpropagation

  • πŸš€ GPU Accelerated: Leverage PyTorch’s GPU support for fast computation

  • πŸ“Š Multiple Methods: Seven differentiation algorithms for different use cases

  • πŸ“ˆ Higher-Order Derivatives: Support for 2nd-order and multi-order derivative computation

  • πŸ”§ Easy API: Simple functional and object-oriented interfaces

  • πŸ§ͺ Well Tested: Validated against the reference derivative package

Quick Example

import torch
import torch_dxdt

t = torch.linspace(0, 2 * torch.pi, 100)
x = torch.sin(t) + 0.1 * torch.randn(100)

# Compute derivative
dx = torch_dxdt.dxdt(x, t, kind="savitzky_golay", window_length=11, polyorder=3)

Indices and tables