Skip to contents

Generate a Graphviz DOT Diagram for NMF-AR / NMF-VAR Models

Source: R/nmfkc.ar.R
nmfkc.ar.DOT.Rd

Produces a Graphviz DOT script for visualizing autoregressive NMF-with-covariates models constructed via nmfkc.ar + nmfkc.

The diagram displays three types of directed relationships:

  • Lagged predictors: \(T_{t-k} \rightarrow X\),

  • Current latent factors: \(X \rightarrow T_t\),

  • Optional intercept effects: Const -> X.

Importantly, no direct edges from lagged variables to current outputs (\(T_{t-k} \rightarrow T_t\)) are drawn, in accordance with the NMF-AR formulation.

Each block of lagged variables is displayed in its own DOT subgraph (e.g., “T-1”, “T-2”, ...), while latent factor nodes and current-time outputs are arranged in separate clusters.

Usage

nmfkc.ar.DOT(
  result,
  degree = 1,
  intercept = any(colnames(result$C) == "(Intercept)"),
  threshold = 0.1,
  rankdir = "RL",
  fill = TRUE,
  weight_scale_xy = 5,
  weight_scale_lag = 5,
  weight_scale_int = 3,
  hide.isolated = TRUE
)

Arguments

result

A fitted nmfkc object representing the AR model. Must contain matrices X and C.

degree

Maximum AR lag to visualize.

intercept

Logical; if TRUE, draws intercept nodes for columns named "(Intercept)" in matrix C. Default is TRUE when an intercept column is detected in C, FALSE otherwise (auto-detected).

threshold

Minimum coefficient magnitude required to draw an edge.

rankdir

Graphviz rank direction (e.g., "RL", "LR", "TB").

fill

Logical; whether nodes are filled with color.

weight_scale_xy

Scaling factor for edges \(X \rightarrow T\).

weight_scale_lag

Scaling factor for lagged edges \(T-k \rightarrow X\).

weight_scale_int

Scaling factor for intercept edges.

hide.isolated

Logical. If TRUE (default), Y nodes that have no edges at or above threshold are excluded from the graph.

Value

A character string representing a Graphviz DOT file.

See also

nmfkc.ar, nmfkc, plot.nmfkc.DOT

Examples

d <- AirPassengers
ar_data <- nmfkc.ar(d, degree = 2)
result <- nmfkc(ar_data$Y, ar_data$A, rank = 1)
#> Y(1,142)~X(1,1)C(1,3)A(3,142)=XB(1,142)...
#> 0sec
dot <- nmfkc.ar.DOT(result, degree = 2)
cat(dot)
#> digraph NMF_AR {
#>   graph [rankdir=RL compound=true];
#>   splines=true; nodesep=0.4; ranksep=0.7; fontname="Arial";
#> 
#>   // Current-time outputs (T)
#>   subgraph cluster_Y{label="T" style="rounded" color="black" penwidth=1.0;
#>   node [shape=box, style="filled,rounded", fillcolor="lightblue", color=black, penwidth=1.5];
#>     1 [label="1"];
#>   }
#> 
#>   // Latent variables (X)
#>   subgraph cluster_X{label="Latent Variables" style="rounded" color="black" penwidth=1.0;
#>   node [shape=ellipse, style="filled,rounded", fillcolor="wheat", color=black, penwidth=1.0];
#>     Basis1 [label="Basis1"];
#>   }
#> 
#>   edge [fontname="Arial", fontsize=8, arrowhead=open];
#> 
#>   // X -> T edges (factor loadings)
#>   edge [color="gray0", fontcolor="gray0", style=solid];
#>   Basis1 -> 1 [label="1.0", penwidth=5.00];
#> 
#>   // Intercept nodes
#>   Const1 [shape=circle, label="11.0"];
#>   Const1 -> Basis1 [penwidth=3.00];
#> 
#>   // Lag block T-1
#>   subgraph cluster_C1 {label="T-1" style="rounded";
#>     node [shape=box, style="filled,rounded", fillcolor="lightcoral", color=black, penwidth=1.5];
#>     1_1 [label="1"];
#>   }
#>   // T-1 -> X edges
#>   edge [color=black, fontcolor=black, style=solid];
#>   1_1 -> Basis1 [label="0.9", penwidth=5.00];
#> }