Parametric t-SNE

---

This research was conducted by the students André de Macedo Wlodkowski @andrewlod and Kalebe Rodrigues Szlachta @kalebers, mentored by professor Jean Paul Barddal @jpbarddal at Pontifical Catholic University of Paraná for the Computer Science graduation final project.

ENGLISH: Archives referring to the prototype are inside the folder parametric-tsne-keras. To execute, it's necessary to follow the software requirements presented on chapter 4 of the artefact.

Main archives:

• TSNEClassifier.ipynb: file that contains all experiments reffering to the research and implementation of the TSNEClassifier class. Executed on Anaconda environment with Jupyter lab commands.

• parametric_tsne.py: main file for Parametric t-SNE implementation, by Luke Lee.

---

t-distributed stochastic neighbor embedding, abbreviated as t-SNE, provides the novel method to apply non-linear dimensionality reduction technique that preserves the local structure of original dataset. However, in order to transform newly prepared points, a model must be re-trained with whole dataset. This would be extremely inefficient provided that our previous dataset describes the plausible distribution already. Parametric t-SNE instead gives you an explicit mapping between original data and the embedded points. It is achieved by building a parametric model for prediction and training it using the same loss as t-SNE.

Getting Started

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes. See deployment for notes on how to deploy the project on a live system.

Prerequisites

This program was tested under Python 3.6. All necessary packages are contained inside requirements.txt.

Installing

After cloning this repository, install required packages by running the following:

pip3 install -r requirements.txt

Running the Tests

parametric_tsne.py can be run directly from command-line. See help for details.

python3 parametric_tsne.py -h

Deployment

Simply create ParametricTSNE instance. The interface was designed similarly to that of scikit-learn estimators.

from parametric_tsne import ParametricTSNE

transformer = ParametricTSNE()

# suppose you have the dataset X
X_new = transformer.fit_transform(X)

# transform new dataset X2 with pre-trained model
X2_new = transformer.transform(X2)

Built With

• scikit-learn - Extensive machine learning framework

• Keras - Deep learning framework wrapper that supports TensorFlow, Theano, and CNTK

Authors

• Luke Lee - Research and implementation - luke0201

Acknowledgements

• This project was forked from zaburo-ch's implementation.