biswajitsc / matrixsample

Matrix Sampling Algorithms.

MatrixSample

Matrix Sampling Algorithms

Updates:

• All recent work is being done in the svm_multiclass folder.
• svm_multiclass/Experiments_Data_BinarySVM.csv contains the test errors for all the methods. where NT denotes no change in datamatrix and M-NT denotes Mean test error on NT method and V-NT denotes variance of test error on NT method. Similarly CW denotes Clarkson Woodruff method and M-CW and V-CW denote the mean and variance in CW method respectively.
• For now methods SS and ABSS have not been run. So, M-SS and M-ABSS have been kept 100.0.
• svm_multiclass/Experiments_Data_Norm.csv contains the error in the norm of the original vector and the projected vector. Here also the header denote the same thing as in svm_multiclass/Experiments_Data_BinarySVM.csv. Also, SS and ABSS have not been run and kept at 100.0 for now.
• PlotWrapper_Norm() is used for obtaining svm_multiclass/Experiments_Data_Norm.csv. It calls experiments_norm.m which calls find_norm_error.m. Fix these 3 files if you want to change some thing for norm error things.

matrixsample's Issues

Generate Plots.

Generate plots from svm_multiclass/Experiments_Data_BinarySVM.csv and svm_multiclass/Experiments_Data_Norm.csv in the following way.

1. Vary N keeping D same.
2. Vary D keeping N same.

GenData2.m

Problem here is n << D. So rank of S < n, So. when we are choosing 0.8*d as subspace, we are getting error.

SS and ABSS not running.

The methods SS and ABSS and taking too long to run. Please run them and get the results both for classification error and norm error using PlotWrapper,m and PlotWrapper_Norm.m respecively.