Summary method for nmf.ffb objects

Produces a formatted summary of a fitted NMF-FFB model, including matrix dimensions, convergence, stability diagnostics, fit statistics, and inference results (if available). For nmf.ffb(method = "fiml") objects the fit statistics are the log-likelihoods, parameter counts and BIC of the feed-forward null, the unpenalized feedback fit and the BIC-selected model, the two likelihood-ratio statistics, and – after nmf.ffb.inference – their bootstrap p-values and the false-selection rate under the null.

Usage

# S3 method for class 'nmf.ffb'
summary(object, ...)

Arguments

object

An object of class "nmf.ffb" returned by nmf.ffb.

...

Not used.

Value

An object of class "summary.nmf.ffb" (the fitted model tagged for printing); printed by print.summary.nmf.ffb.

See also

nmf.ffb, nmf.ffb.inference

Examples

Y <- t(iris[, -5])
Y1 <- Y[1:2, ]; Y2 <- Y[3:4, ]
result <- nmf.ffb(Y1, Y2, rank = 2, maxit = 500)
summary(result)
#> NMF-FFB (FIML, X fixed from stage 1): Y1(2,N) = X(2,2) [C1(2,2) Y1 + C2(2,2) Y2]
#> Stage 1 NMF | stage 2 FIML: 391 / 500 (converged)  epsilon = 1e-06  fiml 91 / 3000
#> 
#> Stability diagnostics:
#>   Spectral radius(XC1): 0.0000 (stable)
#>   ||XC1||_1:            0.0000
#> 
#> Likelihood (FF null | unpenalized feedback | BIC-selected):
#>   loglik:     -346.165 |   -346.165 |   -346.165
#>   npar:              9 |         11 |          9
#>   BIC:          737.43 |     747.45 |     737.43
#>   LR vs null: full = 0.000 (free entries = 2), selected = 0.000 (nnz = 0)
#>   selected: C1.L1 = Inf, nnz = 0, rho(XC1) = 0.0000, restriction = union
#> 
#>   Support candidates (best 2 of 2 by BIC; one parameter costs log(N) = 5.01):
#>     nnz=0  rho=0.000 BIC=      737.43 (selected)  feed-forward (no feedback)
#>     nnz=2  rho=0.000 BIC=      747.45 (+10.02)    Factor2<-Sepal.Length, Factor1<-Sepal.Width
#> 
#> Fit statistics:
#>   MAE (mean absolute error):       1.3106
#>   Effective Rank:                  1.58 / 2  (78.9%)