torchid 0.3.0

GPU-accelerated intrinsic dimension estimators in PyTorch (port of scikit-dimension).

torchid

GPU-accelerated intrinsic dimension estimators in PyTorch. A port of scikit-dimension with batched/vectorized implementations and CUDA support.

Status: in-progress port. See CHECKLIST.md for progress.

Why

scikit-dimension is the reference library for intrinsic dimension (ID) estimation but is CPU-only and relies heavily on per-point Python loops. torchid re-implements every estimator using batched torch ops so the same methods run 10–200× faster on GPU while producing outputs that match the reference library within documented tolerances.

Install

pip install torchid

For a CUDA-capable install, pick the PyTorch wheel that matches your driver, e.g.:

pip install torch --index-url https://download.pytorch.org/whl/cu128
pip install torchid

For running parity tests against scikit-dimension from a clone:

uv sync --group validation

Usage

import torch
from torchid.estimators import lPCA

X = torch.randn(10_000, 50, device="cuda")
est = lPCA().fit(X)
print(est.dimension_)