Skip to content

3D T-SNE graphs with sliders and checkboxes to visualize the T-SNE cloud at every epoch for specific labels. Optionally you can also track specific datapoint by labeling it with a unique marker.

License

Notifications You must be signed in to change notification settings

krishnachaitanya7/Manifolk

Repository files navigation

Manifolk

Did you think that wouldn't it be nice to have a tool that would plot T-SNE graphs at each epoch, and only of specific labels you want to visualize? Which can help you visualize the T-SNE graph for specific labels de-cluttering the T-SNE cloud with many other labels? Find out which datapoints are near to the cluster and which aren't? Look no more!

Manifolk is a Plotly+Dash+MySQLite web application that does exactly that. During your model training process, you generate T-SNE embeddings using your favourite package (e.g. sklearn's T-SNE) and then use Manifolk's API to store all of them into an SQLite database. After the training is over the Dash app can be launched to visualize the 3D T-SNE embeddings. beware the T-SNE embeddings you generate must be 3D, not 2D (because why not have an extra dimension?).

Installation Instructions

pip install --upgrade git+https://github.com/krishnachaitanya7/Manifolk.git

Python API

Screenshots

Changing Epochs

The below screenshot shows how to change epochs and visualize the T-SNE at every epoch

Changing Labels

The below screenshot shows how to change labels

Marking a unique datapoint

The below image shows how to mark a unique datapoint which you want to track across epochs and different labels in the TSNE cloud

Use Cases

Run Instructions

To run the TSNE-GUI plot:

cd $CURRENT_PROJECT_WORKSPACE
python tsne_plot.py

About

3D T-SNE graphs with sliders and checkboxes to visualize the T-SNE cloud at every epoch for specific labels. Optionally you can also track specific datapoint by labeling it with a unique marker.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published