shapescience/tsne-streaming

Implementation of the tSNE embedding enabling streaming visualization.

Computing tSNE embeddings in streaming

Deprecated: the npm packages needed are broken. Use this project only as inspiration 🙂

For more details read the blog post. As explained, at the moment we only have "fake" streaming. Stay tuned.

TODO

• Learn directly mappers: see Learning a Parametric Embedding by Preserving Local Structure).
• Do real streaming like in A-tSNE.

Server

Install

For less trouble, setup python3 with anaconda. Missing packages can be installed with 'pip install X Y Z'

Configuration

To use with your own data you'll need to define in data.py:

• how to load your data samples. We offer digits by default.
• which pipeline you use for feature engineering.

Then you may want to change learners used in embedding.py. We use SVR at the moment. To do a seach for decent calibration run:

python embedding.py

Note that this tool can we used for real-time update of regression results, whatever that means.

Run

python server.py

Client

Have nodejs ready, and it should be as simple as:

cd app
npm install
npm start

If you want to change ports, etc, modify the .webpack.config files.