paradime 0.2.0

A framework for parametric dimensionality reduction

paraDime: A Framework for Parametric Dimensionality Reduction

paraDime is a modular framework for specifying and training parametric dimensionality reduction (DR) models. These models allow you to add new data points to existing low-dimensional representations of high-dimensional data. ParaDime DR models are constructed from simple building blocks (such as the relations between data points), so that experimentation with novel DR techniques becomes easy.

Here you can see a parametric version of t-SNE [1] trained on a subset of 5000 images of handwritten digits from the MNIST dataset [2]:

The rest of the 60,000 images can then be easily embedded into the same space without retraining the t-SNE:

Installation

paraDime is available via PyPi through:

pip install paradime

paraDime requires Numpy, SciPy, scikit-learn, PyNNDescent, and PyTorch (see requirements.txt file).

In order to use PyTorch’s CUDA functionality, it might be necessary to install PyTorch separately with the correct setting for the cudatoolkit option (assuming you have the CUDA Toolkit already installed). See the PyTorch docs for installation info.

Getting Started

For a simple exaple with one of the predefined paraDime routines, see Getting Started in the documentation.

More detailed information about how to set up cusom routines can be found in Building Blocks of a paraDime Routine.

References

[1]

Van Der Maaten, L., Hinton, G. “Visualizing data using t-SNE”, Journal of Machine Learning Research (2008).

[2]

LeCun, Y., Cortes, C., Burges, C.J.C. “The MNIST database of handwritten digits” (1998).