![]() |
planc
Parallel Lowrank Approximation with Non-negativity Constraints
|
#include <distntfnes.hpp>
Public Member Functions | |
DistNTFNES (const Tensor &i_tensor, const int i_k, algotype i_algo, const UVEC &i_global_dims, const UVEC &i_local_dims, const UVEC &i_nls_sizes, const UVEC &i_nls_idxs, const NTFMPICommunicator &i_mpicomm) | |
~DistNTFNES () | |
void | num_iterations (const int i_n) |
Returns number of iterations. More... | |
size_t | modes () const |
Returns the numbers of modes of the tensor. More... | |
size_t | rank () const |
Low Rank. More... | |
void | regularizers (const FVEC i_regs) |
L1 and L2 Regularization for every mode. More... | |
void | compute_error (bool i_error) |
Sets whether to compute the error or not. More... | |
VEC | lambda () |
Returns the lambda of the NCP factors. More... | |
int | current_it () const |
Returns the current outer iteration of the computeNTF. More... | |
double | current_error () const |
Returns the current error. More... | |
void | dim_tree (bool i_dim_tree) |
MTTKRP can be computed with or without dimension trees. More... | |
void | accelerated (const bool &set_acceleration) |
Does the algorithm need acceleration? More... | |
bool | is_stale_mttkrp (const int ¤t_mode) const |
void | reset (const NCPFactors &new_factors, bool trans=false) |
This function will completely reset all the factors and the state of AUNTF. More... | |
void | factor (int mode, double *factor_matrix) |
Returns the factor matrix by collected it across all the processors. More... | |
void | computeNTF () |
The main computeNTF loop. More... | |
Definition at line 10 of file distntfnes.hpp.
|
inline |
Definition at line 202 of file distntfnes.hpp.
|
inline |
Definition at line 230 of file distntfnes.hpp.
|
inlineinherited |
Does the algorithm need acceleration?
Definition at line 513 of file distauntf.hpp.
|
inlineinherited |
Sets whether to compute the error or not.
Definition at line 488 of file distauntf.hpp.
|
inlineinherited |
The main computeNTF loop.
Definition at line 578 of file distauntf.hpp.
|
inlineinherited |
Returns the current error.
Definition at line 498 of file distauntf.hpp.
|
inlineinherited |
Returns the current outer iteration of the computeNTF.
Definition at line 496 of file distauntf.hpp.
|
inlineinherited |
MTTKRP can be computed with or without dimension trees.
Dimtree is default.
Definition at line 501 of file distauntf.hpp.
|
inlineinherited |
Returns the factor matrix by collected it across all the processors.
Don't call this during the computeNTF loop. It is right now called after all the iterations to dump the factor matrix.
[in] | mode | of the factor matrix to be collected across processors |
[in] | factor_matrix. | To be allocated and freed by the caller. |
Definition at line 552 of file distauntf.hpp.
|
inlineinherited |
Definition at line 518 of file distauntf.hpp.
|
inlineinherited |
Returns the lambda of the NCP factors.
Definition at line 494 of file distauntf.hpp.
|
inlineinherited |
Returns the numbers of modes of the tensor.
Definition at line 482 of file distauntf.hpp.
|
inlineinherited |
Returns number of iterations.
Definition at line 480 of file distauntf.hpp.
|
inlineinherited |
Low Rank.
Definition at line 484 of file distauntf.hpp.
|
inlineinherited |
L1 and L2 Regularization for every mode.
Definition at line 486 of file distauntf.hpp.
|
inlineinherited |
This function will completely reset all the factors and the state of AUNTF.
Use it with caution!!! It is the responsibility of the caller to preserve the current factor states and other relevant computations
Definition at line 529 of file distauntf.hpp.