Personalized PCA

This is the implementation of the paper Personalized PCA: decoupling shared and unique features.

If you run into any problems, please submit an issue, or contact the authors of that paper.

How to test a simple example

• run python3 ppca.py --dataset=borrowpowertest --logoutput=True

Results will be generated to the "outputs/" folder

How to run the video segmentation experiment

• Create a folder called "images/"
• Put all the image frames of the video in the "images/" folder
• Rewrite the first line of the function load_car_data in the file imgpro.py accordingly
• Run python3 ppca.py --dataset='img_test' --logoutput=True
• If needed, fine-tune 'ngc' and 'nlc' to get the best results

How to run the presidential debate topics modeling experiment

• Download the dataset from this link
• Put all files in the "debate/" folder
• Run python3 ppca.py --dataset='debate_test' --logoutput=True

File organization

• ppca.py contains all the hyperparameters
• algs.py contains the implementation of personalized PCA learning algorithm
• imgpro.py, vectorize.py, mnist.py are used to handle video data, debate corpus, and FEMNIST dataset respectively.