synsNMF {musclesyneRgies} | R Documentation |
Non-negative matrix factorisation
Description
Non-negative matrix factorisation
Usage
synsNMF(
V,
R2_target = 0.01,
runs = 5,
max_iter = 1000,
last_iter = 20,
MSE_min = 1e-04,
fixed_syns = NA
)
Arguments
V |
EMG data frame to be reconstructed, usually filtered and time-normalised |
R2_target |
Threshold to stop iterations for a certain factorisation rank |
runs |
Number of repetitions for each rank to avoid local minima |
max_iter |
Maximum number of iterations allowed for each rank |
last_iter |
How many of the last iterations should be checked before stopping? |
MSE_min |
Threshold on the mean squared error to choose the factorisation rank or minimum number of synergies |
fixed_syns |
To impose the factorisation rank or number of synergies |
Details
The first column of V
must always contain time information.
Value
Object of class musclesyneRgies
with elements:
-
syns
factorisation rank or minimum number of synergies
-
M
motor modules (time-invariant coefficients)
-
P
motor primitives (time-dependent coefficients)
-
V
original data
-
Vr
reconstructed data
-
iterations
number of iterations to convergence
-
R2
quality of reconstruction (coefficient of determination) -
rank_type
was the rankfixed
orvariable
?
-
classification
classification type (e.g.,none
,k-means
,NMF
, etc.)
References
Lee, D. D. & Seung, H. S.
Learning the parts of objects by non-negative matrix factorization.
Nature 401, 788-91 (1999).
Santuz, A., Ekizos, A., Janshen, L., Baltzopoulos, V. & Arampatzis, A.
On the Methodological Implications of Extracting Muscle Synergies from Human Locomotion.
Int. J. Neural Syst. 27, 1750007 (2017).
FĂ©votte, C., Idier, J. Algorithms for Nonnegative Matrix Factorization with the Beta-Divergence Neural Computation 23, 9 (2011).
Examples
# Note that for bigger data sets one might want to run computation in parallel
# Load some data
data(FILT_EMG)
# Extract synergies (careful, rank is imposed here!)
SYNS <- lapply(FILT_EMG, synsNMF, fixed_syns = 4)