Hinge GAN with orthonormal generator for robust sparse PCA under Huber's contamination model.
To run the hinge GAN, a cuda gpu is required.
python train.py --r 20 --k 200 --p 2000 --n 1000 --eps 0.1
For the ITSPCA and RegSPCA, run the following script on cpu:
python spca_reg_ita.py --r 20 --k 200 --p 2000 --n 1000 --eps 0
For both scripts, the output is the estimation error of the principal subspace projection matrix measured in squared Frobenius norm. To obtain the estimated parameters, access the model
object after evaluation, as it stores the estimated V_hat as attributes V_hat, V_hat_ITSPCA, and V_hat_RegSPCA.