Package jline.api.infer

Class InferLqn

java.lang.Object

jline.api.infer.InferLqn

public class InferLqn
extends Object

LQN parameter identification via an Extended Kalman Filter.

Estimates hidden Layered Queueing Network parameters (activity host demands, task think times) from measured performance data (response times, utilizations, throughputs), following Zheng, Yang, Woodside, Litoiu, Iszlai, "Tracking Time-Varying Parameters in Software Systems with Extended Kalman Filters", CASCON 2005 (equations 1-9). The parameter is modelled as a zero-mean random walk a_k = a_{k-1} + w and the measurement as z_k = h(a_k)+v, where h is the LQN performance model. This is the JAR counterpart of the MATLAB infer_lqn family and the native-Python line_solver.inference.infer_lqn; results match across codebases.

Copyright (c) 2012-2026, Imperial College London. All rights reserved.

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static class	InferLqn.JacobianResult	Finite-difference sensitivity matrix H = dh/da together with h0 = h(a).

Method Summary

Modifier and Type	Method	Description
static InferLqnResult	ekf(Function<Matrix,Matrix> hfun, Matrix a0, Matrix P0, Matrix Z, Matrix Q, Matrix R, InferLqnOptions opt)	Extended Kalman Filter tracking a hidden parameter vector across Z.
static Matrix	getObs(LayeredNetworkAvgTable table, List<ObsSpec> obsSpec)	Extract the observation vector selected by obsSpec from a solved table.
static Matrix	getParams(LayeredNetwork model, List<ParamSpec> paramSpec)	Read the current values of the parameters named in paramSpec.
static InferLqnResult	inferLqn(LayeredNetwork model, List<ParamSpec> paramSpec, List<ObsSpec> obsSpec, Matrix Z, InferLqnOptions opt)	Identify hidden LQN parameters from measured performance data.
static InferLqn.JacobianResult	jacobian(Function<Matrix,Matrix> hfun, Matrix a, double fdStep, double fdFloor)	Forward finite-difference sensitivity matrix of h at a.
static void	setParams(LayeredNetwork model, List<ParamSpec> paramSpec, Matrix a)	Apply a parameter vector to a LayeredNetwork in place.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Method Details

setParams

public static void setParams(LayeredNetwork model,
 List<ParamSpec> paramSpec,
 Matrix a)

Apply a parameter vector to a LayeredNetwork in place.

Parameters:

model - the LQN whose parameters are set

paramSpec - list of parameters to set

a - (np x 1) parameter vector

getParams

public static Matrix getParams(LayeredNetwork model,
 List<ParamSpec> paramSpec)

Read the current values of the parameters named in paramSpec.

getObs

public static Matrix getObs(LayeredNetworkAvgTable table,
 List<ObsSpec> obsSpec)

Extract the observation vector selected by obsSpec from a solved table.

jacobian

public static InferLqn.JacobianResult jacobian(Function<Matrix,Matrix> hfun,
 Matrix a,
 double fdStep,
 double fdFloor)

Forward finite-difference sensitivity matrix of h at a.

Column i is (h(a + d_i) - h0)/d_i with d_i = fdStep*max(|a_i|, fdFloor). hfun is evaluated np+1 times. This is the approximate sensitivity matrix H_k used in the EKF update.

ekf

public static InferLqnResult ekf(Function<Matrix,Matrix> hfun,
 Matrix a0,
 Matrix P0,
 Matrix Z,
 Matrix Q,
 Matrix R,
 InferLqnOptions opt)

Extended Kalman Filter tracking a hidden parameter vector across Z.

Parameters:

hfun - maps a parameter vector to a predicted observation z = h(a)

a0 - (np x 1) initial estimate

P0 - (np x np) initial covariance

Z - (no x nsteps) measurement matrix, one column per step

Q - (np x np) drift covariance

R - (no x no) measurement covariance

opt - options (fdStep, fdFloor, clampPositive, aTrue, verbose)

inferLqn

public static InferLqnResult inferLqn(LayeredNetwork model,
 List<ParamSpec> paramSpec,
 List<ObsSpec> obsSpec,
 Matrix Z,
 InferLqnOptions opt)

Identify hidden LQN parameters from measured performance data.

Estimates the parameters named in paramSpec (activity host demands and/or task think times) of the LayeredNetwork model from the measurement sequence Z (no x nsteps) using an EKF over the observation model defined by obsSpec. A single measurement column with opt.QFac = 0 reduces to one-shot least-squares calibration. On return the model has the final estimate applied.