-
Notifications
You must be signed in to change notification settings - Fork 55
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Merge pull request #452 from sandialabs/remove-unused-function-defs
Remove commented-out function definitions and delete files with no executable code
- Loading branch information
Showing
91 changed files
with
832 additions
and
9,037 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -3295,80 +3295,81 @@ def symmetric_low_rank_spectrum_update(update, orig_e, U, proj_U, force_rank_inc | |
| #return the new eigenvalues | ||
| return new_evals, True | ||
|
|
||
| #Note: This function won't work for our purposes because of the assumptions | ||
| #about the rank of the update on the nullspace of the matrix we're updating, | ||
| #but keeping this here commented for future reference. | ||
| #Function for doing fast calculation of the updated inverse trace: | ||
| #def riedel_style_inverse_trace(update, orig_e, U, proj_U, force_rank_increase=True): | ||
| # """ | ||
| # input: | ||
| # | ||
| # update : ndarray | ||
| # symmetric low-rank update to perform. | ||
| # This is the first half the symmetric rank decomposition s.t. | ||
| # [email protected]= the full update matrix. | ||
| # | ||
| # orig_e : ndarray | ||
| # Spectrum of the original matrix. This is a 1-D array. | ||
| # | ||
| # proj_U : ndarray | ||
| # Projector onto the complement of the column space of the | ||
| # original matrix's eigenvectors. | ||
| # | ||
| # output: | ||
| # | ||
| # trace : float | ||
| # Value of the trace of the updated psuedoinverse matrix. | ||
| # | ||
| # updated_rank : int | ||
| # total rank of the updated matrix. | ||
| # | ||
| # rank_increase_flag : bool | ||
| # a flag that is returned to indicate is a candidate germ failed to amplify additional parameters. | ||
| # This indicates things short circuited and so the scoring function should skip this germ. | ||
| # """ | ||
| # | ||
| # #First we need to for the matrix P, whose column space | ||
| # #forms an orthonormal basis for the component of update | ||
| # #that is in the complement of U. | ||
| # | ||
| # proj_update= proj_U@update | ||
| # | ||
| # #Next take the RRQR decomposition of this matrix: | ||
| # q_update, r_update, _ = _sla.qr(proj_update, mode='economic', pivoting=True) | ||
| # | ||
| # #Construct P by taking the columns of q_update corresponding to non-zero values of r_A on the diagonal. | ||
| # nonzero_indices_update= _np.nonzero(_np.diag(r_update)>1e-10) #HARDCODED (threshold is hardcoded) | ||
| # | ||
| # #if the rank doesn't increase then we can't use the Riedel approach. | ||
| # #Abort early and return a flag to indicate the rank did not increase. | ||
| # if len(nonzero_indices_update[0])==0 and force_rank_increase: | ||
| # return None, None, False | ||
| # | ||
| # P= q_update[: , nonzero_indices_update[0]] | ||
| # | ||
| # updated_rank= len(orig_e)+ len(nonzero_indices_update[0]) | ||
| # | ||
| # #Now form the matrix R_update which is given by P.T @ proj_update. | ||
| # R_update= P.T@proj_update | ||
| # | ||
| # #R_update gets concatenated with U.T@update to form | ||
| # #a block column matrixblock_column= np.concatenate([U.T@update, R_update], axis=0) | ||
| # | ||
| # Uta= U.T@update | ||
| # | ||
| # try: | ||
| # RRRDinv= R_update@_np.linalg.inv(R_update.T@R_update) | ||
| # except _np.linalg.LinAlgError as err: | ||
| # print('Numpy thinks this matrix is singular, condition number is: ', _np.linalg.cond(R_update.T@R_update)) | ||
| # print((R_update.T@R_update).shape) | ||
| # raise err | ||
| # pinv_orig_e_mat= _np.diag(1/orig_e) | ||
| # | ||
| # trace= _np.sum(1/orig_e) + _np.trace( RRRDinv@(_np.eye(Uta.shape[1]) + Uta.T@pinv_orig_e_mat@Uta)@RRRDinv.T ) | ||
| # | ||
| # return trace, updated_rank, True | ||
| # Note: Th function below won't work for our purposes because of the assumptions | ||
| # about the rank of the update on the nullspace of the matrix we're updating, | ||
| # but keeping this here commented for future reference. | ||
| ''' | ||
| def riedel_style_inverse_trace(update, orig_e, U, proj_U, force_rank_increase=True): | ||
| """ | ||
| input: | ||
| update : ndarray | ||
| symmetric low-rank update to perform. | ||
| This is the first half the symmetric rank decomposition s.t. | ||
| [email protected]= the full update matrix. | ||
| orig_e : ndarray | ||
| Spectrum of the original matrix. This is a 1-D array. | ||
| proj_U : ndarray | ||
| Projector onto the complement of the column space of the | ||
| original matrix's eigenvectors. | ||
| output: | ||
| trace : float | ||
| Value of the trace of the updated psuedoinverse matrix. | ||
| updated_rank : int | ||
| total rank of the updated matrix. | ||
| rank_increase_flag : bool | ||
| a flag that is returned to indicate is a candidate germ failed to amplify additional parameters. | ||
| This indicates things short circuited and so the scoring function should skip this germ. | ||
| """ | ||
| #First we need to for the matrix P, whose column space | ||
| #forms an orthonormal basis for the component of update | ||
| #that is in the complement of U. | ||
| proj_update= proj_U@update | ||
| #Next take the RRQR decomposition of this matrix: | ||
| q_update, r_update, _ = _sla.qr(proj_update, mode='economic', pivoting=True) | ||
| #Construct P by taking the columns of q_update corresponding to non-zero values of r_A on the diagonal. | ||
| nonzero_indices_update= _np.nonzero(_np.diag(r_update)>1e-10) #HARDCODED (threshold is hardcoded) | ||
| #if the rank doesn't increase then we can't use the Riedel approach. | ||
| #Abort early and return a flag to indicate the rank did not increase. | ||
| if len(nonzero_indices_update[0])==0 and force_rank_increase: | ||
| return None, None, False | ||
| P= q_update[: , nonzero_indices_update[0]] | ||
| updated_rank= len(orig_e)+ len(nonzero_indices_update[0]) | ||
| #Now form the matrix R_update which is given by P.T @ proj_update. | ||
| R_update= P.T@proj_update | ||
| #R_update gets concatenated with U.T@update to form | ||
| #a block column matrixblock_column= np.concatenate([U.T@update, R_update], axis=0) | ||
| Uta= U.T@update | ||
| try: | ||
| RRRDinv= R_update@_np.linalg.inv(R_update.T@R_update) | ||
| except _np.linalg.LinAlgError as err: | ||
| print('Numpy thinks this matrix is singular, condition number is: ', _np.linalg.cond(R_update.T@R_update)) | ||
| print((R_update.T@R_update).shape) | ||
| raise err | ||
| pinv_orig_e_mat= _np.diag(1/orig_e) | ||
| trace= _np.sum(1/orig_e) + _np.trace( RRRDinv@(_np.eye(Uta.shape[1]) + Uta.T@pinv_orig_e_mat@Uta)@RRRDinv.T ) | ||
| return trace, updated_rank, True | ||
| ''' | ||
|
|
||
| def minamide_style_inverse_trace(update, orig_e, U, proj_U, force_rank_increase=False): | ||
| """ | ||
| This function performs a low-rank update to the components of | ||
|
|
||
Oops, something went wrong.