OnlineNMF.jl (Julia API)

Non-negative Matrix Factorization (NMF)

OnlineNMF.nmf — Method

nmf(;input::AbstractString="", outdir::Union{Nothing,AbstractString}=nothing, alpha::Number=1, beta::Number=2, graphv::Number=0, l1u::Number=eps(Float32), l1v::Number=eps(Float32), l2u::Number=eps(Float32), l2v::Number=eps(Float32), dim::Number=3, numepoch::Number=5, chunksize::Number=1, algorithm::AbstractString="frobenius", lower::Number=0, upper::Number=1.0f+38, initU::Union{Nothing,AbstractString}=nothing, initV::Union{Nothing,AbstractString}=nothing, initL::Union{Nothing,AbstractString}=nothing, logdir::Union{Nothing,AbstractString}=nothing)

Non-negative Matrix Factorization (NMF).

Input Arguments

input : Julia Binary file generated by OnlinePCA.csv2bin function.
outdir : The directory user want to save the result.
alpha : The parameter of Alpha-divergence.
beta : The parameter of Beta-divergence.
graphv : Graph regularization parameter for the factor matrix V.
l1u : L1-regularization parameter for the factor matrix U.
l1v : L1-regularization parameter for the factor matrix V.
l2u : L2-regularization parameter for the factor matrix U.
l2v : L2-regularization parameter for the factor matrix V.
dim : The number of dimension of NMF.
numepoch : The number of epochs.
`chunksize: The number of rows reading at once (e.g. 5000).
algorithm: Update Algorithm. frobenius, kl, is, pearson, hellinger, neyman, alpha, and beta are available.
lower : Stopping Criteria (When the relative change of error is below this value, the calculation is terminated).
upper : Stopping Criteria (When the relative change of error is above this value, the calculation is terminated).
initU : The CSV file saving the initial values of factor matrix U.
initV : The CSV file saving the initial values of factor matrix V.
initL : The CSV file saving the initial values of graph laplacian L.
logdir : The directory where intermediate files are saved in every epoch.

Output Arguments

U : Factor matrix (No. rows of the data matrix × dim)
V : Factor matrix (No. columns of the data matrix × dim)
stop : Whether the calculation is converged

source

Discretized Non-negative Matrix Factorization (DNMF)

OnlineNMF.dnmf — Method

dnmf(;input::AbstractString="", outdir::Union{Nothing,AbstractString}=nothing, beta::Number=2, binu::Number=eps(Float32), binv::Number=eps(Float32), teru::Number=eps(Float32), terv::Number=eps(Float32), graphv::Number=0, l1u::Number=eps(Float32), l1v::Number=eps(Float32), l2u::Number=eps(Float32), l2v::Number=eps(Float32), dim::Number=3, numepoch::Number=5, chunksize::Number=1, lower::Number=0, upper::Number=1.0f+38, initU::Union{Nothing,AbstractString}=nothing, initV::Union{Nothing,AbstractString}=nothing, initL::Union{Nothing,AbstractString}=nothing, logdir::Union{Nothing,AbstractString}=nothing)

Discretized Non-negative Matrix Factorization (DNMF).

Input Arguments

input : Julia Binary file generated by OnlinePCA.mm2bin function.
outdir : The directory user want to save the result.
beta : The parameter of Beta-divergence.
binu : Binary {0,1} regularization parameter for the factor matrix U.
binv : Binary {0,1} regularization parameter for the factor matrix V.
teru : Ternary {0,1,2} regularization parameter for the factor matrix U.
terv : Ternary {0,1,2} regularization parameter for the factor matrix V.
graphv : Graph regularization parameter for the factor matrix V.
l1u : L1-regularization parameter for the factor matrix U.
l1v : L1-regularization parameter for the factor matrix V.
l2u : L2-regularization parameter for the factor matrix U.
l2v : L2-regularization parameter for the factor matrix V.
dim : The number of dimension of NMF.
numepoch : The number of epochs.
`chunksize: The number of rows reading at once (e.g. 5000).
lower : Stopping Criteria (When the relative change of error is below this value, the calculation is terminated)
upper : Stopping Criteria (When the relative change of error is above this value, the calculation is terminated)
initU : The CSV file saving the initial values of factor matrix U.
initV : The CSV file saving the initial values of factor matrix V.
initL : The CSV file saving the initial values of graph laplacian L.
logdir : The directory where intermediate files are saved, in every epoch.

Output Arguments

U : Factor matrix (No. rows of the data matrix × dim)
V : Factor matrix (No. columns of the data matrix × dim)
stop : Whether the calculation is converged

source

Sparse Non-negative Matrix Factorization (SNMF)

OnlineNMF.sparse_nmf — Method

sparse_nmf(;input::AbstractString="", outdir::Union{Nothing,AbstractString}=nothing, alpha::Number=1, beta::Number=2, graphv::Number=0, l1u::Number=eps(Float32), l1v::Number=eps(Float32), l2u::Number=eps(Float32), l2v::Number=eps(Float32), dim::Number=3, numepoch::Number=5, chunksize::Number=1, algorithm::AbstractString="frobenius", lower::Number=0, upper::Number=1.0f+38, initU::Union{Nothing,AbstractString}=nothing, initV::Union{Nothing,AbstractString}=nothing, initL::Union{Nothing,AbstractString}=nothing, logdir::Union{Nothing,AbstractString}=nothing)

Sparse Non-negative Matrix Factorization (SNMF).

Input Arguments

input : Julia Binary file generated by OnlinePCA.mm2bin function.
outdir : The directory user want to save the result.
alpha : The parameter of Alpha-divergence.
beta : The parameter of Beta-divergence.
graphv : Graph regularization parameter for the factor matrix V.
l1u : L1-regularization parameter for the factor matrix U.
l1v : L1-regularization parameter for the factor matrix V.
l2u : L2-regularization parameter for the factor matrix U.
l2v : L2-regularization parameter for the factor matrix V.
dim : The number of dimension of NMF.
numepoch : The number of epochs.
`chunksize: The number of rows reading at once (e.g. 5000).
algorithm: Update Algorithm. frobenius, kl, is, pearson, hellinger, neyman, alpha, and beta are available.
lower : Stopping Criteria (When the relative change of error is below this value, the calculation is terminated).
upper : Stopping Criteria (When the relative change of error is above this value, the calculation is terminated).
initU : The CSV file saving the initial values of factor matrix U.
initV : The CSV file saving the initial values of factor matrix V.
initL : The CSV file saving the initial values of graph laplacian L.
logdir : The directory where intermediate files are saved in every epoch.

Output Arguments

U : Factor matrix (No. rows of the data matrix × dim)
V : Factor matrix (No. columns of the data matrix × dim)
stop : Whether the calculation is converged

source

Sparse Discretized Non-negative Matrix Factorization (SDNMF)

OnlineNMF.sparse_dnmf — Method

sparse_dnmf(;input::AbstractString="", outdir::Union{Nothing,AbstractString}=nothing, beta::Number=2, binu::Number=eps(Float32), binv::Number=eps(Float32), teru::Number=eps(Float32), terv::Number=eps(Float32), graphv::Number=0, l1u::Number=eps(Float32), l1v::Number=eps(Float32), l2u::Number=eps(Float32), l2v::Number=eps(Float32), dim::Number=3, numepoch::Number=5, chunksize::Number=1, lower::Number=0, upper::Number=1.0f+38, initU::Union{Nothing,AbstractString}=nothing, initV::Union{Nothing,AbstractString}=nothing, initL::Union{Nothing,AbstractString}=nothing, logdir::Union{Nothing,AbstractString}=nothing)

Sparse Discretized Non-negative Matrix Factorization (SDNMF).

Input Arguments

input : Julia Binary file generated by OnlinePCA.mm2bin function.
outdir : The directory user want to save the result.
beta : The parameter of Beta-divergence.
binu : Binary {0,1} regularization parameter for the factor matrix U.
binv : Binary {0,1} regularization parameter for the factor matrix V.
teru : Ternary {0,1,2} regularization parameter for the factor matrix U.
terv : Ternary {0,1,2} regularization parameter for the factor matrix V.
graphv : Graph regularization parameter for the factor matrix V.
l1u : L1-regularization parameter for the factor matrix U.
l1v : L1-regularization parameter for the factor matrix V.
l2u : L2-regularization parameter for the factor matrix U.
l2v : L2-regularization parameter for the factor matrix V.
dim : The number of dimension of NMF.
numepoch : The number of epochs.
`chunksize: The number of rows reading at once (e.g. 5000).
lower : Stopping Criteria (When the relative change of error is below this value, the calculation is terminated)
upper : Stopping Criteria (When the relative change of error is above this value, the calculation is terminated)
initU : The CSV file saving the initial values of factor matrix U.
initV : The CSV file saving the initial values of factor matrix V.
initL : The CSV file saving the initial values of graph laplacian L.
logdir : The directory where intermediate files are saved, in every epoch.

Output Arguments

U : Factor matrix (No. rows of the data matrix × dim)
V : Factor matrix (No. columns of the data matrix × dim)
stop : Whether the calculation is converged

source

BinCOO Non-negative Matrix Factorization (BinCOONMF)

OnlineNMF.bincoo_nmf — Method

bincoo_nmf(;input::AbstractString="", outdir::Union{Nothing,AbstractString}=nothing, alpha::Number=1, beta::Number=2, graphv::Number=0, l1u::Number=eps(Float32), l1v::Number=eps(Float32), l2u::Number=eps(Float32), l2v::Number=eps(Float32), dim::Number=3, numepoch::Number=5, chunksize::Number=1, algorithm::AbstractString="frobenius", lower::Number=0, upper::Number=1.0f+38, initU::Union{Nothing,AbstractString}=nothing, initV::Union{Nothing,AbstractString}=nothing, initL::Union{Nothing,AbstractString}=nothing, logdir::Union{Nothing,AbstractString}=nothing)

Sparse Non-negative Matrix Factorization (SNMF).

Input Arguments

input : Julia Binary file generated by OnlinePCA.mm2bin function.
outdir : The directory user want to save the result.
alpha : The parameter of Alpha-divergence.
beta : The parameter of Beta-divergence.
graphv : Graph regularization parameter for the factor matrix V.
l1u : L1-regularization parameter for the factor matrix U.
l1v : L1-regularization parameter for the factor matrix V.
l2u : L2-regularization parameter for the factor matrix U.
l2v : L2-regularization parameter for the factor matrix V.
dim : The number of dimension of NMF.
numepoch : The number of epochs.
`chunksize: The number of rows reading at once (e.g. 5000).
algorithm: Update Algorithm. frobenius, kl, is, pearson, hellinger, neyman, alpha, and beta are available.
lower : Stopping Criteria (When the relative change of error is below this value, the calculation is terminated).
upper : Stopping Criteria (When the relative change of error is above this value, the calculation is terminated).
initU : The CSV file saving the initial values of factor matrix U.
initV : The CSV file saving the initial values of factor matrix V.
initL : The CSV file saving the initial values of graph laplacian L.
logdir : The directory where intermediate files are saved in every epoch.

Output Arguments

U : Factor matrix (No. rows of the data matrix × dim)
V : Factor matrix (No. columns of the data matrix × dim)
stop : Whether the calculation is converged

source

BinCOO Discretized Non-negative Matrix Factorization (BinCOODNMF)

OnlineNMF.bincoo_dnmf — Method

bincoo_dnmf(;input::AbstractString="", outdir::Union{Nothing,AbstractString}=nothing, beta::Number=2, binu::Number=eps(Float32), binv::Number=eps(Float32), teru::Number=eps(Float32), terv::Number=eps(Float32), graphv::Number=0, l1u::Number=eps(Float32), l1v::Number=eps(Float32), l2u::Number=eps(Float32), l2v::Number=eps(Float32), dim::Number=3, numepoch::Number=5, chunksize::Number=1, lower::Number=0, upper::Number=1.0f+38, initU::Union{Nothing,AbstractString}=nothing, initV::Union{Nothing,AbstractString}=nothing, initL::Union{Nothing,AbstractString}=nothing, logdir::Union{Nothing,AbstractString}=nothing)

Sparse Discretized Non-negative Matrix Factorization (SDNMF).

Input Arguments

input : Julia Binary file generated by OnlinePCA.mm2bin function.
outdir : The directory user want to save the result.
beta : The parameter of Beta-divergence.
binu : Binary {0,1} regularization parameter for the factor matrix U.
binv : Binary {0,1} regularization parameter for the factor matrix V.
teru : Ternary {0,1,2} regularization parameter for the factor matrix U.
terv : Ternary {0,1,2} regularization parameter for the factor matrix V.
graphv : Graph regularization parameter for the factor matrix V.
l1u : L1-regularization parameter for the factor matrix U.
l1v : L1-regularization parameter for the factor matrix V.
l2u : L2-regularization parameter for the factor matrix U.
l2v : L2-regularization parameter for the factor matrix V.
dim : The number of dimension of NMF.
numepoch : The number of epochs.
`chunksize: The number of rows reading at once (e.g. 5000).
lower : Stopping Criteria (When the relative change of error is below this value, the calculation is terminated)
upper : Stopping Criteria (When the relative change of error is above this value, the calculation is terminated)
initU : The CSV file saving the initial values of factor matrix U.
initV : The CSV file saving the initial values of factor matrix V.
initL : The CSV file saving the initial values of graph laplacian L.
logdir : The directory where intermediate files are saved, in every epoch.

Output Arguments

U : Factor matrix (No. rows of the data matrix × dim)
V : Factor matrix (No. columns of the data matrix × dim)
stop : Whether the calculation is converged

source

Graph Laplacian (L = D - A, where D is a degree matrix and A is an adjacent matrix).

OnlineNMF.graph_laplacian — Function

graph_laplacian(A::AbstractArray, norm::Bool=true)

Graph Laplacian (L = D - A, where D is a degree matrix and A is an adjacent matrix).

Input Arguments

A : A symmetric numerical matrix.
norm : Whether to normalize the graph Laplacian or not.

Output Arguments

Graph Laplacian Matrix (No. rows/columns of A × No. rows/columns of A)

source