Compute class means → compute overall mean → compute within-class scatter → compute between-class scatter → solve generalized eigenvalue problem → sort eigenvalues → form projection matrix → project data.

Compute eigenvectors first → compute class means → compute overall mean → form projection matrix → compute scatter matrices → project data → sort eigenvalues.

Compute the overall mean → compute class means → compute within-class scatter → compute between-class scatter → solve generalized eigenvalue problem → sort eigenvalues → form projection matrix → project data.

Compute within-class scatter → compute between-class scatter → compute class means → compute overall mean → solve eigenvalue problem → sort eigenvalues → project data → form projection matrix.