computeFunctionValue

Method of CPNTensor.

Unified CPN function-value primitive: evaluates any block of equations at an operating point.

Source: src/tensor/@CPNTensor/computeFunctionValue.m

Syntax

f = obj.computeFunctionValue(sys, v)
f = obj.computeFunctionValue(sys, v, equationRows, columnIndex)
f = obj.computeFunctionValue(sys, v, equationRows, columnIndex, mode)

Description

computeFunctionValue evaluates the multilinear function stored in the tensor at the full signal vector v and returns the values of the selected equations,

$f = \Phi(\text{equationRows},\text{columnIndex}) \; m\bigl(S(:,\text{columnIndex}), v\bigr),$

where $m$ is the vector of monomial column products and $\Phi$ the parameter matrix.

The per-factor formula — True $\to v$ , False $\to 1-v$ , Continuous $c \to c\,v + (1-|c|)(1-b\,v)$ — is selected by $b=$ sys.mtiBase, monomial or literal; see CPNTensor and mss.

The tensor is read from sys.mtiTensor. v must be in structure-matrix row order, that is, already scattered through the owning model’s index vectors; assembling it is the caller’s job.

Evaluation is delegated to the MTISIM kernel. The tensor is first compiled by prepareForMTISIM, which bakes sys.mtiBase into its output, so the kernel itself is base-agnostic.

Modes

mode chooses how the parameter matrix is applied. Both settings give identical results and differ only in where the work happens:

"scatter" (default) passes $\Phi$ straight into MTISIM, which scatters the column products through it in one call — no external sparse mat-vec.
"passthrough" has MTISIM return the raw column products and applies $\Phi$ afterwards in MATLAB.

Both handle non-square blocks: the output row count is decoupled from the [x; u] split, and the whole signal vector is passed as one block with no separate inputs. An unrecognised mode raises CPNTensor:computeFunctionValue:mode.

Input arguments

Argument	Description
`obj`	`CPNTensor`. Ignored — the tensor is taken from `sys.mtiTensor`.
`sys`	Owning `mss` or `mdss` model. Supplies `mtiTensor` and `mtiBase`.
`v`	Full signal vector in structure-matrix row order, assembled by the caller.
`equationRows`	Parameter-matrix rows to return. Default: all equations.
`columnIndex`	Monomial columns to evaluate. Default: all columns.
`mode`	`"scatter"` (default) or `"passthrough"`.

Empty ([]) is accepted for equationRows, columnIndex and mode, and selects that argument’s default.

Output arguments

Output	Description
`f`	Values of the selected equations at `v`, one entry per entry of `equationRows`.

Example

%      x1  x2  x1x2
S = [   1   0    1     % x1
        0   1    1 ];  % x2
P = [ -10  10    0     % dx1 = -10*x1 + 10*x2
        0   0    1 ];  % dx2 = x1*x2

T   = CPNTensor(S,P);
sys = mss(T, uint32([1;2]), uint32(zeros(0,1)), 0, ...
             uint32([1;2]), uint32(zeros(0,1)));
v   = [0.3; 0.7];

T.computeFunctionValue(sys, v)          % [4; 0.21]
T.computeFunctionValue(sys, v, 2, 3)    % 0.21 - second equation only