by beringresearch

beringresearch / ivis

Dimensionality reduction in very large datasets using Siamese Networks

210 Stars 19 Forks Last release: 5 months ago (1.8.1) GNU General Public License v2.0 420 Commits 22 Releases

Available items

No Items, yet!

The developer of this repository has not created any items for sale yet. Need a bug fixed? Help with integration? A different license? Create a request here:

DOI DOI Documentation Status Downloads Build Status


Implementation of the ivis algorithm as described in the paper Structure-preserving visualisation of high dimensional single-cell datasets. Ivis is designed to reduce dimensionality of very large datasets using a siamese neural network trained on triplets. Both unsupervised and supervised modes are supported.

ivis 10M data points


Ivis runs on top of TensorFlow. To install the latest ivis release from PyPi running on the CPU TensorFlow package, run:

# TensorFlow 2 packages require a pip version >19.0.
pip install --upgrade pip
pip install ivis[cpu]

If you have CUDA installed and want ivis to use the tensorflow-gpu package, run

pip install ivis[gpu]

Development version can be installed directly from from github:

git clone https://github.com/beringresearch/ivis
cd ivis
pip install -e '.[cpu]'

The following optional dependencies are needed if using the visualization callbacks while training the Ivis model: - matplotlib - seaborn


Ivis Python package is updated frequently! To upgrade, run:

pip install ivis --upgrade


  • Scalable: ivis is fast and easily extends to millions of observations and thousands of features.
  • Versatile: numpy arrays, sparse matrices, and hdf5 files are supported out of the box. Additionally, both categorical and continuous features are handled well, making it easy to apply ivis to heterogeneous problems including clustering and anomaly detection.
  • Accurate: ivis excels at preserving both local and global features of a dataset. Often, ivis performs better at preserving global structure of the data than t-SNE, making it easy to visualise and interpret high-dimensional datasets.
  • Generalisable: ivis supports addition of new data points to original embeddings via a
    method, making it easy to incorporate ivis into standard sklearn Pipelines.

And many more! See ivis readme for latest additions and examples.


from ivis import Ivis
from sklearn.preprocessing import MinMaxScaler
from sklearn import datasets

iris = datasets.load_iris() X = iris.data X_scaled = MinMaxScaler().fit_transform(X)

model = Ivis(embedding_dims=2, k=15)

embeddings = model.fit_transform(X_scaled)

Ivis Universe

Ivis can be used in a wide variety of real-world applications. The Ivis Universe consists of packages that extend the core Ivis functionality.

  • ivis-animate - visualise the Ivis learning process.
  • ivis-explain - explain which features contribute the most to ivis embeddings.

Copyright 2020 Bering Limited

We use cookies. If you continue to browse the site, you agree to the use of cookies. For more information on our use of cookies please see our Privacy Policy.