patrick-kidger/torchcde

Differentiable controlled differential equation solvers for PyTorch with GPU support and memory-efficient adjoint backpropagation.

55
/ 100
Established

This tool helps machine learning practitioners build models that analyze and make predictions from irregular time series data, like sensor readings or financial transactions. It takes your raw, potentially messy time-series data and transforms it into a continuous signal, then uses that to train a "Neural Controlled Differential Equation" model. The output is a highly accurate model for tasks like classification or forecasting, even with missing values or unevenly spaced data.

475 stars. Used by 1 other package. No commits in the last 6 months. Available on PyPI.

Use this if you need to build state-of-the-art models for complex, irregular time series data where traditional RNNs struggle with gaps or varied sampling rates.

Not ideal if you are starting a new project, as a more performant and production-ready tool called Diffrax is now recommended by the creators of this library.

time-series-analysis machine-learning-engineering predictive-modeling data-science neural-networks
Stale 6m
Maintenance 2 / 25
Adoption 11 / 25
Maturity 25 / 25
Community 17 / 25

How are scores calculated?

Stars

475

Forks

50

Language

Python

License

Apache-2.0

Last pushed

Sep 04, 2025

Commits (30d)

0

Dependencies

3

Reverse dependents

1

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/patrick-kidger/torchcde"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.