Multiway-Algorithms PARAFAC Comparison Test

To compare the multiway-algorithms PARAFAC implementation with the matlab nway toolbox run the following step:

1. Initialize the mutliway-algorithms and nway-toolbox local installations by running

   $ bash init.sh

   This should clone the latest multiway-algorithms develop branch and the nway repository

2. Run the nway PARAFAC implementation:

   $ bash run-nway-toolbox-parafac.sh

   Results in:

   Loss =
   
     413.6739
   

3. Run the multiway-algorithms PARAFAC implementation:

   $ bash run-multiway-alg-parafac.sh

   Results in:

   [INFO ] 18:14:11.388 [Main.main()] Main - Loss = 413.9942821962213
   

The data can be found in matlab/Fluorescence\ EEMs/ as .dat files or in data/ as .csv files. The matlab script matlab/generate_data_dir.m generates the .csv files.

Algorithm Setup

Both runs are started with the same options:

Option	Value
number of components	4..10
max. iterations	2500
improvement tolerance threshold	10e-6
init method	{random orthogonalized matrices, SVD}

Results

number of components	nway loss (random orth. init)	mwa loss (random orth. init)	nway loss (SVD init)	mwa loss (SVD init)
3	562.3193	562.6590	562.3179	562.9468
4	413.7502	413.9942	413.7559	414.0267
5	306.6297	307.1157	306.691	310.7814
6	217.8061	218.3507	217.7974	218.0926
7	175.0781	175.6058	175.0551	175.9703
8	144.0486	145.9342	143.9903	145.5471
9	122.1815	126.5388	122.2394	122.3786
10	104.1776	109.5957	104.1796	107.6224