Project 6 Visualization

Purpose of this project:

• Gain a basic appreciation of data science programming and data visualization
• Implement a data science algorithm with given pseudocode
• Be aware of efficiency of the implemented code

In this project we implemented three different methods of visualizing data; PCA, Isomap and t-SNE.

The data we are trying to visualize are the following: Swiss Roll and OptDigits:

Swiss Roll looks like this:

Whilst OptDigits is a dataset where there are 64 features, and is thus not possible to visualize without transforming first.

Running the project

Python 3+ is required to run the project.
Install the required dependencies by running pip install -r requirements.txt

To run the three different visualization algorithms, run the main.py file.

The results should be something like this:

Results of PCA

Swiss Roll:

We see that the Swiss Roll is reduced to two dimensions such that the curl is visualized.

OptDigits:

We see that applying PCA to OptDigits yields poor results.

Results of Isomap

Swiss Roll:

We see that the Swiss Roll is reduced to two dimensions such that the roll is 'Rolled out'.

OptDigits:

We see that applying PCA to OptDigits is better than PCA, but it still fails to distinguish the different clusters.

Results of t-SNE on OptiDigits

Here we can see that applying t-SNE to the OptiDigits dataset clearly distinguishes the different clusters.