### Simulate Nominal Model with Deterministic Noise

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Simulates the nominal plant model with a stationary input and deterministic noise. This setup is used to observe parameter drift due to insufficient excitation, matching Rohr's example.

```julia
simulate(nominal_sim!, nominal_ic;
    input_signal = 2.0,
    deterministic_noise = (x, p, t) -> 0.5sin(16.1t),
    noise_param = nothing,
    tspan = (0.0, 100.0),
    adapt_gain = 3.35)
```

--------------------------------

### Training with LBFGS Optimizer

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/DiffEqFlux

Continues training the model using the LBFGS optimizer, starting from the results obtained with ADAM.

```julia
res2 = sciml_train(loss_n_ode, res1.minimizer, LBFGS(); cb)
cb(res2.minimizer, loss_n_ode(res2.minimizer)...; doplot = true)
```

--------------------------------

### Get Val-Wrapped Keys for ComponentVector

Source: https://docs.sciml.ai/ComponentArrays/stable/api

Returns Val-wrapped keys of a ComponentVector for fast iteration over component keys. This method also works directly on an AbstractAxis.

```julia
julia> using ComponentArrays

julia> ca = ComponentArray(a = 1, b = [1, 2, 3], c = (a = 4,))
ComponentVector{Int64}(a = 1, b = [1, 2, 3], c = (a = 4))

julia> [ca[k] for k in valkeys(ca)]
3-element Vector{Any}:
 1
  [1, 2, 3]
  ComponentVector{Int64}(a = 4)

julia> sum(prod(ca[k]) for k in valkeys(ca))
11
```

--------------------------------

### ComponentArrays.labels

Source: https://docs.sciml.ai/ComponentArrays/stable/api

Get string labels for each index of a ComponentVector. This is useful for automatic plot legend labeling.

```APIDOC
## labels(x::ComponentVector)

### Description
Get string labels for for each index of a `ComponentVector`. Useful for automatic plot legend labelling.

### Parameters
#### Path Parameters
- **x** (ComponentVector) - The ComponentVector to get labels from.

### Response
#### Success Response
- Returns a `Vector{String}` containing the labels for each index.
```

--------------------------------

### Get Axes of a ComponentArray

Source: https://docs.sciml.ai/ComponentArrays/stable/api

Demonstrates using the `getaxes` function to retrieve the axes associated with a ComponentArray, which differs from the standard `axes` function.

```julia
using ComponentArrays

ax = Axis(a = 1:3, b = (4:6, (a = 1, b = 2:3)))

A = zeros(6, 6)

ca = ComponentArray(A, (ax, ax))

getaxes(ca)
```

--------------------------------

### Get String Labels for ComponentVector Indices

Source: https://docs.sciml.ai/ComponentArrays/stable/api

Retrieves string labels for each index of a ComponentVector. This is useful for automatic plot legend labeling.

```julia
julia> x = ComponentArray(a = 5, b = [(a = (a = 20, b = 1), b = 0), (a = (a = 33, b = 1), b = 0)],
           c = (a = (a = 2, b = [1, 2]), b = [1.0 2.0; 5 6]))
ComponentVector{Float64}(a = 5.0, b = ComponentVector{Float64, SubArray{Float64, 1, Vector{Float64}, Tuple{UnitRange{Int64}}, true}, Tuple{Axis{(a = ViewAxis(1:2, Axis(a = 1, b = 2)), b = 3)}}}[(a = (a = 20.0, b = 1.0), b = 0.0), (a = (a = 33.0, b = 1.0), b = 0.0)], c = (a = (a = 2.0, b = [1.0, 2.0]), b = [1.0 2.0; 5.0 6.0]))

julia> ComponentArrays.labels(x)
14-element Vector{String}:
 "a"
 "b[1].a.a"
 "b[1].a.b"
 "b[1].b"
 "b[2].a.a"
 "b[2].a.b"
 "b[2].b"
 "c.a.a"
 "c.a.b[1]"
 "c.a.b[2]"
 "c.b[1,1]"
 "c.b[2,1]"
 "c.b[1,2]"
 "c.b[2,2]"
```

--------------------------------

### Define Simulation Inputs

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Sets up simulation time span, input signal function, and initial conditions for various components.

```julia
# Simulation time span
tspan = (0.0, 30.0)

# Input signal
input_signal = (x, p, t) -> sin(3t)

# Initial conditions
ref_ic = zeros(1)
nominal_ic = zeros(1)
truth_ic = zeros(3)
sensor_ic = zeros(1)
θ_est_ic = ComponentArray(θr = 0.0, θy = 0.0)
```

--------------------------------

### Setting Up the ODE Problem and Data

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/DiffEqFlux

Defines the initial conditions, time span, the true ODE function, and generates synthetic data for training.

```julia
u0 = Float32[2.0; 0.0]
datasize = 30
tspan = (0.0f0, 1.5f0)

function trueODEfunc(du, u, p, t)
    true_A = [-0.1 2.0; -2.0 -0.1]
    du .= ((u .^ 3)'true_A)'
end

t = range(tspan[1], tspan[2], length = datasize)
prob = ODEProblem(trueODEfunc, u0, tspan)
ode_data = Array(solve(prob, Tsit5(), saveat = t))
```

--------------------------------

### Set Up and Simulate Adaptive Control System

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Defines the simulation function, including noise handling, parameter estimation, and solving ODE/SDE problems. Use this function to run adaptive control simulations.

```julia
function simulate(plant_fun, plant_ic;
        tspan = tspan,
        input_signal = input_signal,
        adapt_gain = 1.5,
        noise_param = nothing,
        deterministic_noise = 0.0)
    noise(D, vars, p, t) = (D.feedback_loop.sensor[1] = noise_param)

    # Truth control parameters
    θ_truth = (r = bm/bp, y = (ap-am)/bp)

    # Initial conditions
    ic = ComponentArray(
        reference_model = ref_ic,
        feedback_loop = (
            parameter_estimates = θ_est_ic,
            sensor = sensor_ic,
            plant_model = plant_ic
        )
    )

    # Model parameters
    p = (; gamma = adapt_gain, plant_fun)

    sim_fun = apply_inputs(system!; r = input_signal, n = deterministic_noise)

    # We can also choose whether we want to include random noise in our model by switching between an ODE
    # and an SDE problem.
    if noise_param === nothing
        prob = ODEProblem(sim_fun, ic, tspan, p, max_iters = 2000)
    else
        prob = SDEProblem(sim_fun, noise, ic, tspan, p, max_iters = 2000)
    end

    ## Solve!
    sol = solve(prob)

    ## Plot
    # Reference model tracking
    top = plot(
        sol,
        vars = ["reference_model[1]", "feedback_loop.sensor"],
        legend = :right,
        title = "Reference Model Tracking"
    )

    # Parameter estimate tracking
    bottom = plot(sol, vars = "feedback_loop.parameter_estimates")
    plot!(
        bottom,
        [tspan...], [θ_truth.r θ_truth.y; θ_truth.r θ_truth.y],
        labels = ["θr truth" "θy truth"],
        legend = :right,
        title = "Parameter Estimate Tracking"
    )

    # Combine both plots
    plot(top, bottom, layout = (2, 1), size = (800, 800))
end
```

--------------------------------

### Full System Simulation

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Simulates the complete system, including the reference model and the feedback loop. Takes input signal `r` and noise `n`.

```julia
function system!(D, vars, p, t; r = 0.0, n = 0.0)
    @unpack reference_model, feedback_loop = vars

    ym = ref_sim!(D.reference_model, reference_model, (), t; u = r)
    yp = feedback_sys!(D.feedback_loop, feedback_loop, p, t; ym, r, n)
    return yp
end
```

--------------------------------

### Interactive GUI for system simulation

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/coulomb_control

An interactive GUI allowing users to adjust PID gains (kp, ki, kd), switch between Coulomb and Viscous friction models, and select input types (Sine or Step) with varying magnitude and period. It visualizes the system's response and phase plot.

```julia
## Interactive GUI for switching out plant models and varying PID gains
@manipulate for kp in 0:0.01:15,
    ki in 0:0.01:15,
    kd in 0:0.01:15,
    damping in Dict(
        "Coulomb" => coulomb_block!,
        "Viscous" => viscous_block!
    ),
    reference in Dict(
        "Sine" => sine_input,
        "Step" => step_input
    ),
    magnitude in 0:0.01:10, # pop-pop!
    period in 1:0.01:30,
    plot_v in false

    # Inputs
    tspan = (0.0, 30.0)

    ctrl_fun = PID_controller!
    # plant_fun = coulomb_block!

    ref = if reference==sine_input
        reference(; period, mag = magnitude)
    else
        reference(mag = magnitude)
    end

    m = 50.0
    μ = 0.1
    ω = 2π/period
    c = 4*μ*m*g/(π*ω*magnitude) # Viscous equivalent damping
    k = 50.0

    plant_p = (; m, μ, c, k) # We'll just put everything for both models in here
    ctrl_p = (; kp, ki, kd)

    plant_ic = (v = 0, x = 0)
    ctrl_ic = (; x = 0)

    # Set up and solve
    sys_p = (
        ctrl = (params = ctrl_p, fun = ctrl_fun),
        plant = (params = plant_p, fun = damping)
    )
    sys_ic = ComponentArray(ctrl = ctrl_ic, plant = plant_ic)
    sys_fun = ODEFunction(simulator(feedback_sys!; ref), syms = [:u, :v, :x])
    sys_prob = ODEProblem(sys_fun, sys_ic, tspan, sys_p)

    sol = solve(sys_prob, Tsit5())

    # Plot
    t = tspan[1]:0.1:tspan[2]
    lims = magnitude*[-1, 1]
    plotvars = plot_v ? [3, 2] : [3]
    strip = plot(t, ref.(0, 0, t), ylim = 1.2lims, label = "r(t)")
    plot!(strip, sol, vars = plotvars)
    phase = plot(ref.(0, 0, t), map(x->x.plant.x, sol(t).u),
        xlim = lims,
        ylim = 1.2lims,
        legend = false,
        xlabel = "r(t)",
        ylabel = "x(t)"
    )
    plot(strip, phase, layout = (2, 1), size = (700, 800))
end
```

--------------------------------

### Importing Necessary Libraries

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/DiffEqFlux

Imports essential libraries for numerical computation, differential equations, plotting, and optimization.

```julia
using ComponentArrays
using OrdinaryDiffEq
using Plots
using UnPack

using DiffEqFlux: sciml_train
using Flux: glorot_uniform, ADAM
using Optim: LBFGS
```

--------------------------------

### Importing Necessary Libraries

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Imports essential Julia packages for component-based arrays, control systems, differential equations, unpacking, and plotting.

```julia
using ComponentArrays
using ControlSystems
using DifferentialEquations
using UnPack
using Plots
```

--------------------------------

### Simulate with Unmodeled Dynamics

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Simulates a system with unmodeled dynamics, demonstrating the robustness of adaptive control. This is useful for testing how well the adaptation law performs when the plant model is not perfectly known.

```julia
simulate(truth_sim!, truth_ic; noise_param = 0.2)
```

--------------------------------

### Indexing with Multiple Symbols to Select and Reorder Components

Source: https://docs.sciml.ai/ComponentArrays/stable/indexing_behavior

Demonstrates indexing a ComponentArray with a tuple of symbols to select specific components and change their order.

```julia
ca[(:c, :a)]
```

--------------------------------

### Simulate Truth Model with Deterministic Noise

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Simulates the truth plant model with a stationary input and deterministic noise. This demonstrates how parameter drift, caused by insufficient excitation, can lead to system instability.

```julia
simulate(truth_sim!, truth_ic;
    input_signal = 2.0,
    deterministic_noise = (x, p, t) -> 0.5sin(16.1t),
    noise_param = nothing,
    tspan = (0.0, 72.9),
    adapt_gain = 3.35)
```

--------------------------------

### Open-loop response of Coulomb friction system

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/coulomb_control

Simulates and plots the open-loop response of the sliding block with Coulomb friction when a constant input of 5 is applied. Requires defining constants for time span, mass, friction coefficient, and stiffness.

```julia
const tspan = (0.0, 30.0)
const m = 50.0
const μ = 0.1
const k = 50.0

p = (; m, μ, k)
ic = ComponentArray(v = 0, x = 0)

ODEProblem(simulator(coulomb_block!, u = 5), ic, tspan, p) |> solve |> plot
```

--------------------------------

### Applying Inputs with Keyword Arguments

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Defines a utility function `apply_inputs` that allows system inputs to be specified as values or functions of (x, p, t) via keyword arguments. This is useful for integrating control systems into differential equation solvers.

```julia
maybe_apply(f::Function, x, p, t) = f(x, p, t)
maybe_apply(f, x, p, t) = f

function apply_inputs(func; kwargs...)
    simfun(dx, x, p, t) = func(
        dx, x, p, t; map(f->maybe_apply(f, x, p, t), (; kwargs...))...)
    simfun(x, p, t) = func(x, p, t; map(f->maybe_apply(f, x, p, t), (; kwargs...))...)
    return simfun
end
```

--------------------------------

### Training with ADAM Optimizer

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/DiffEqFlux

Sets up a callback for visualization and trains the neural ODE model using the ADAM optimizer for a specified number of iterations.

```julia
cb = function (θ, loss, pred; doplot = false)
    display(loss)
    # plot current prediction against data
    pl = scatter(t, ode_data[1, :], label = "data")
    scatter!(pl, t, pred[1, :], label = "prediction")
    display(plot(pl))
    return false
end
cb(θ, loss_n_ode(θ)...)

data = Iterators.repeated((), 1000)

res1 = sciml_train(loss_n_ode, θ, ADAM(0.05); cb, maxiters = 100)
cb(res1.minimizer, loss_n_ode(res1.minimizer)...; doplot = true)
```

--------------------------------

### Simulate and Plot Step Response

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Generate and plot the step response for both the truth plant and the nominal plant to compare their dynamics. Requires `truth_ic` and `nominal_ic` to be defined.

```julia
step_p = plot(solve(ODEProblem(apply_inputs(truth_sim!; u = 1), truth_ic, (0.0, 10.0)));
    vars = 1, label = "truth model")
plot!(step_p, solve(ODEProblem(apply_inputs(nominal_sim!; u = 1), nominal_ic, (0.0, 10.0)));
    vars = 1, label = "nominal model")
```

--------------------------------

### Run Simulation with Noise

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Executes the simulation function with specified plant model and initial conditions, including deterministic noise. This is used to generate simulation results for analysis.

```julia
simulate(nominal_sim!, nominal_ic; noise_param = 0.2)
```

--------------------------------

### Using @view Macro for Slicing Behavior

Source: https://docs.sciml.ai/ComponentArrays/stable/indexing_behavior

Shows how to use the `@view` macro to ensure that indexing operations create views rather than slices, allowing modifications to affect the original array.

```julia
@view(ca[:b])[1] = 0
ca
```

--------------------------------

### Define Reference Model

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Create a reference model transfer function and its simulator. This defines the desired system behavior.

```julia
am = 3
bm = 3
ref_model = bm / (s + am)
ref_sim! = SISO_simulator(ref_model)
```

--------------------------------

### ComponentArrays.PartitionedAxis

Source: https://docs.sciml.ai/ComponentArrays/stable/api

Defines an Axis for creating arrays of ComponentArrays, specifying a partition size and an index map.

```APIDOC
## `ComponentArrays.PartitionedAxis` — Type

### Description

Axis for creating arrays of `ComponentArray`s.

### Usage

```julia
pa = PartitionedAxis(partition_size, index_map)
```
```

--------------------------------

### Constructing and Modifying ComponentArrays

Source: https://docs.sciml.ai/ComponentArrays/stable/quickstart

Demonstrates the creation of 1D ComponentArrays using NamedTuple-like syntax and shows how to modify individual elements and nested fields. Also shows how to collect the array into a flat vector.

```julia
julia> c = (a = 2, b = [1, 2]);

julia> x = ComponentArray(a = 1.0, b = [2, 1, 4], c = c)
ComponentVector{Float64}(a = 1.0, b = [2.0, 1.0, 4.0], c = (a = 2.0, b = [1.0, 2.0]))

julia> x.c.a = 400;
       x
ComponentVector{Float64}(a = 1.0, b = [2.0, 1.0, 4.0], c = (a = 400.0, b = [1.0, 2.0]))

julia> x[5]
400.0

julia> collect(x)
7-element Vector{Float64}:
   1.0
   2.0
   1.0
   4.0
 400.0
   1.0
   2.0

julia> typeof(similar(x, Int32)) === typeof(ComponentVector{Int32}(a = 1, b = [2, 1, 4], c = c))
true
```

--------------------------------

### Unpacking ComponentArray to StaticArrays

Source: https://docs.sciml.ai/ComponentArrays/stable/static_unpack

Demonstrates the usage of the @static_unpack macro. Use this when you need to perform operations on small array components of a ComponentArray and want to leverage the performance benefits of StaticArrays by avoiding heap allocations.

```julia
julia> x = ComponentVector(a = 5, b = [4, 1], c = [1 2; 3 4], d = (e = 2, f = [6, 30.0]));

julia> @static_unpack a, b, c, d = x;

julia> a
5.0

julia> b
2-element SVector{2, Float64} with indices SOneTo(2):
 4.0
 1.0

julia> c
2×2 SMatrix{2, 2, Float64, 4} with indices SOneTo(2)×SOneTo(2):
 1.0  2.0
 3.0  4.0

julia> d
ComponentVector{Float64,SubArray...}(e = 2.0, f = [6.0, 30.0])
```

--------------------------------

### Creating SISO Simulator from Transfer Function

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Provides a function `SISO_simulator` to create a derivative function for a Single-Input Single-Output (SISO) state-space system, suitable for use with differential equation solvers. It converts a `TransferFunction` or `StateSpace` object into a form where the first state variable represents the system's output.

```julia
SISO_simulator(P::TransferFunction) = SISO_simulator(ss(P))
function SISO_simulator(P::AbstractStateSpace)
    @unpack A, B, C, D = P

    if size(D)!=(1, 1)
        error("This is not a SISO system")
    end

    # Put into transposed observer canonical form so the first element is also the y value
    BB = reverse(vec(C))
    CC = reverse(vec(B))'
    DD = D[1, 1]

    return function sim!(dx, x, p, t; u = 0.0)
        dx .= A*x + BB*u
        return CC*x + DD*u
    end
end
```

--------------------------------

### Closed-loop response with PID control

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/coulomb_control

Simulates and plots the closed-loop response of the system using a PID controller and the Coulomb friction block. Sets up the parameters for both the controller and the plant, and defines the initial conditions.

```julia
p = (
    ctrl = (
        params = (kp = 13, ki = 12, kd = 5),
        fun = PID_controller!
    ),
    plant = (
        params = plant_p,
        fun = coulomb_block!
    )
)

ic = ComponentArray(ctrl = (; x = 0), plant = plant_ic)

sol = ODEProblem(simulator(feedback_sys!, ref = 10), ic, tspan, p) |> solve
plot(sol, vars = 3)
```

--------------------------------

### Feedback System Simulation

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Simulates a feedback control loop, calculating control signals, plant output, tracking error, and updating parameters. Assumes `vars` contains `parameter_estimates`, `plant_model`, and `sensor`, and `p` contains `gamma` and `plant_fun`.

```julia
function feedback_sys!(D, vars, p, t; ym, r, n)
    @unpack parameter_estimates, plant_model, sensor = vars
    γ = p.gamma
    regressor = [r, plant_model[1]]

    u = control(parameter_estimates, regressor)
    yp = p.plant_fun(D.plant_model, plant_model, (), t; u = yp[1])
    ŷ = sensor_sim!(D.sensor, sensor, (), t; u = yp[1]) + n
    e = ŷ .- ym
    regressor[2] = ŷ
    adapt!(D.parameter_estimates, parameter_estimates, γ, t; e, w = regressor)
    return yp
end
```

--------------------------------

### Multi-Symbol Indexing with Tuple vs. Array

Source: https://docs.sciml.ai/ComponentArrays/stable/indexing_behavior

Compares multi-symbol indexing using a tuple of symbols versus an array of symbols, showing they yield equivalent results.

```julia
ca[[:c, :a]] == ca[(:c, :a)]
```

--------------------------------

### Import necessary libraries

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/coulomb_control

Imports required Julia packages for component arrays, differential equations, interactive manipulation, parameter handling, and plotting.

```julia
using ComponentArrays
using DifferentialEquations
using Interact: @manipulate
using Parameters: @unpack
using Plots
```

--------------------------------

### Using @view with Multiple Symbols

Source: https://docs.sciml.ai/ComponentArrays/stable/indexing_behavior

Applies the `@view` macro to multi-symbol indexing to create a view of selected and reordered components.

```julia
@view ca[(:c, :a)]
```

--------------------------------

### Simulate and Plot Sine Response

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Generate and plot the sine response for both the truth plant and the nominal plant. Requires `truth_ic` and `nominal_ic` to be defined.

```julia
u = (x, p, t) -> sin(3t)
sin_p = plot(solve(ODEProblem(apply_inputs(truth_sim!; u), truth_ic, (0.0, 10.0)));
    vars = 1, label = "truth model")
plot!(sin_p, solve(ODEProblem(apply_inputs(nominal_sim!; u), nominal_ic, (0.0, 10.0)));
    vars = 1, label = "nominal model")
```

--------------------------------

### Solve Composed ODE Problem and Unit Test Component

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/ODE_jac

Solves the composed ODE problem and demonstrates how to solve and obtain results for a single component system (Lotka-Volterra) for unit testing purposes.

```julia
## Solve problem
# We can solve the composed system...
comp_sol = solve(comp_prob, Rodas5())

# ...or we can unit test one of the component systems
lotka_sol = solve(lotka_prob, Rodas5())
```

--------------------------------

### Create and Access ComponentArray with Axis

Source: https://docs.sciml.ai/ComponentArrays/stable/api

Demonstrates creating a ComponentArray with a complex Axis structure and accessing its nested components.

```julia
using ComponentArrays

ax = Axis((a = 1, b = ViewAxis(2:7, PartitionedAxis(2, (a = 1, b = 2))), 
           c = ViewAxis(8:10, (a = 1, b = 2:3))))

A = [100, 4, 1.3, 1, 1, 4.4, 0.4, 2, 1, 45]

ca = ComponentArray(A, ax)

ca.a

ca.b

ca.c

ca.c.b
```

--------------------------------

### Input signal generators

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/coulomb_control

Provides functions to generate step and sine wave inputs for the system. These are used to define the reference trajectory.

```julia
step_input(; time = 1.0, mag = 1.0) = (x, p, t) -> t>time ? mag : 0
sine_input(; mag = 1.0, period = 10.0) = (x, p, t) -> mag*sin(t*2π/period)
```

--------------------------------

### Create a ComponentArray

Source: https://docs.sciml.ai/ComponentArrays/stable/indexing_behavior

Initializes a ComponentArray with named scalar and vector components.

```julia
using ComponentArrays

ca = ComponentArray(a = 5, b = [4, 1])
```

--------------------------------

### Construct ComponentArray from NamedTuple or Keyword Arguments

Source: https://docs.sciml.ai/ComponentArrays/stable/api

Shows how to create a ComponentArray using a NamedTuple or keyword arguments, providing struct-like access to elements.

```julia
using ComponentArrays

x = ComponentArray(a = 1, b = [2, 1, 4], c = (a = 2, b = [1, 2]))

x.c.a = 400

x

x[5]

collect(x)
```

--------------------------------

### Constructing ComponentArrays from Existing Arrays

Source: https://docs.sciml.ai/ComponentArrays/stable/quickstart

Shows how to create a new ComponentArray by copying an existing one and overriding or adding fields. Note that nested fields cannot be changed directly during this construction.

```julia
julia> v = ComponentVector(a = 1, b = 2, c = 3);

julia> ComponentVector(v; a = 11, new = 42)
ComponentVector{Int64}(a = 11, b = 2, c = 3, new = 42)
```

--------------------------------

### PID controller implementation

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/coulomb_control

Implements a Proportional-Integral-Derivative (PID) controller. This function calculates the control output based on the error, its integral, and its derivative.

```julia
function PID_controller!(D, vars, p, t; err = 0.0, v = 0.0)
    @unpack kp, ki, kd = p
    @unpack x = vars

    D.x = ki*err
    return x + kp*err + kd*v
end
```

--------------------------------

### Define Nominal Plant Model

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Specify the nominal plant model and its simulator. This represents the known structure for the adaptation law.

```julia
ap = 1
bp = 2
nominal_plant = bp / (s + ap)
nominal_sim! = SISO_simulator(nominal_plant)
```

--------------------------------

### Sliding block with Coulomb friction model

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/coulomb_control

Defines the differential equations for a sliding block system with Coulomb friction. Includes a deadzone to aid simulation stability when velocity and acceleration are near zero.

```julia
# Sliding block with coulomb friction
function coulomb_block!(D, vars, p, t; u = 0.0)
    @unpack m, μ, k = p
    @unpack v, x = vars

    D.x = v
    a = -μ*g*softsign(v) + k*(u-x)/m
    D.v = abs(a)<1e-3 && abs(v)<1e-3 ? -10v : a #deadzone to help the simulation
    return x
end
```

--------------------------------

### Creating a Dense Neural Layer Component

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/DiffEqFlux

Defines a function to create a dense layer component using ComponentArrays, initializing weights and biases.

```julia
dense_layer(in, out) = ComponentArray{Float32}(W = glorot_uniform(out, in), b = zeros(out))
```

--------------------------------

### Define Sensor Dynamics

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Create a first-order sensor dynamics transfer function and its simulator. This is used for adding noise to measurements.

```julia
τ = 0.005
sensor_plant = 1 / (τ*s + 1)
sensor_sim! = SISO_simulator(sensor_plant)
```

--------------------------------

### Define Truth Plant with Unmodeled Dynamics

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Construct the truth plant model by incorporating unmodeled dynamics. This represents the actual system behavior.

```julia
unmodeled_dynamics = 229/(s^2 + 30s + 229)
truth_plant = nominal_plant * unmodeled_dynamics
truth_sim! = SISO_simulator(truth_plant)
```

--------------------------------

### Creating Higher Dimensional ComponentArrays

Source: https://docs.sciml.ai/ComponentArrays/stable/quickstart

Illustrates the creation of a 2D ComponentArray (ComponentMatrix) through broadcasted operations on a 1D ComponentArray. This is useful for automatic dimension expansion in modeling contexts.

```julia
julia> x2 = x .* x'
7×7 ComponentMatrix{Float64} with axes Axis(a = 1, b = ViewAxis(2:4, Shaped1DAxis((3,))), c = ViewAxis(5:7, Axis(a = 1, b = ViewAxis(2:3, Shaped1DAxis((2,)))))) × Axis(a = 1, b = ViewAxis(2:4, Shaped1DAxis((3,))), c = ViewAxis(5:7, Axis(a = 1, b = ViewAxis(2:3, Shaped1DAxis((2,))))))
   1.0    2.0    1.0     4.0     400.0    1.0    2.0
   2.0    4.0    2.0     8.0     800.0    2.0    4.0
   1.0    2.0    1.0     4.0     400.0    1.0    2.0
   4.0    8.0    4.0    16.0    1600.0    4.0    8.0
 400.0  800.0  400.0  1600.0  160000.0  400.0  800.0
   1.0    2.0    1.0     4.0     400.0    1.0    2.0
   2.0    4.0    2.0     8.0     800.0    2.0    4.0

julia> x2[:c, :c]
3×3 ComponentMatrix{Float64} with axes Axis(a = 1, b = ViewAxis(2:3, Shaped1DAxis((2,)))) × Axis(a = 1, b = ViewAxis(2:3, Shaped1DAxis((2,))))
 160000.0  400.0  800.0
    400.0    1.0    2.0
    800.0    2.0    4.0

julia> x2[:a, :a]
1.0

julia> x2[:a, :c]
ComponentVector{Float64}(a = 400.0, b = [1.0, 2.0])

julia> x2[:b, :c]
3×3 ComponentMatrix{Float64} with axes Shaped1DAxis((3,)) × Axis(a = 1, b = ViewAxis(2:3, Shaped1DAxis((2,))))
  800.0  2.0  4.0
  400.0  1.0  2.0
 1600.0  4.0  8.0
```

--------------------------------

### KeepIndex with Integer Index

Source: https://docs.sciml.ai/ComponentArrays/stable/indexing_behavior

Shows how `KeepIndex` can be used with a single integer index to capture a full component and retain its name.

```julia
ca[KeepIndex(1)]
```

--------------------------------

### Defining the Parameter ComponentArray

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/DiffEqFlux

Constructs a ComponentArray to hold the initial conditions and the parameters of the dense neural layers.

```julia
layers = (L1 = dense_layer(2, 50), L2 = dense_layer(50, 2))
θ = ComponentArray(u = u0, p = layers)
```

--------------------------------

### Sliding block with viscous friction model

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/coulomb_control

Defines the differential equations for a sliding block system incorporating viscous friction. Use this for systems where damping is proportional to velocity.

```julia
# Sliding block with viscous friction
function viscous_block!(D, vars, p, t; u = 0.0)
    @unpack m, c, k = p
    @unpack v, x = vars

    D.x = v
    D.v = (-c*v + k*(u-x))/m
    return x
end
```

--------------------------------

### Feedback system definition

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/coulomb_control

Defines the overall feedback system, combining a controller and a plant model. It calculates the control input `u` and applies it to the plant.

```julia
function feedback_sys!(D, components, p, t; ref = 0.0)
    @unpack ctrl, plant = components

    u = p.ctrl.fun(D.ctrl, ctrl, p.ctrl.params, t; err = ref-plant.x, v = -plant.v)
    return p.plant.fun(D.plant, plant, p.plant.params, t; u)
end
```

--------------------------------

### KeepIndex with Full Range Index

Source: https://docs.sciml.ai/ComponentArrays/stable/indexing_behavior

Demonstrates using `KeepIndex` with a range that spans multiple components, retaining the names of the components that are fully captured.

```julia
ca[KeepIndex(1:3)]
```

--------------------------------

### Indexing Behavior: Slicing vs. In-place Update

Source: https://docs.sciml.ai/ComponentArrays/stable/indexing_behavior

Illustrates that indexing with symbols slices the array, creating a copy, while direct assignment to a component updates in-place. This highlights the default slicing behavior of ComponentArrays.

```julia
ca[:b][1] = 0
ca[:a] = 0
ca
```

--------------------------------

### Gradient Descent Adaptation Law

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Implements the gradient descent adaptation law for updating parameters. Requires `e` (error) and `w` (regressor vector) as keyword arguments.

```julia
function adapt!(Dθ, θ, γ, t; e, w)
    Dθ .= -γ*e*w
    return nothing
end
```

--------------------------------

### Create a ComponentArray with Nested Structure

Source: https://docs.sciml.ai/ComponentArrays/stable/indexing_behavior

Initializes a ComponentArray with nested named components, including a nested tuple.

```julia
ca = ComponentArray(a = 5, b = [4, 1], c = (a = 2, b = [6, 30.0]))
```

--------------------------------

### Define simulator utility function

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/coulomb_control

A utility function to wrap simulation functions, allowing for optional application of functions based on their type.

```julia
const g = 9.80665

maybe_apply(f::Function, x, p, t) = f(x, p, t)
maybe_apply(f, x, p, t) = f

# Applies functions of form f(x,p,t) to be applied and passed in as inputs
function simulator(func; kwargs...)
    simfun(dx, x, p, t) = func(
        dx, x, p, t; map(f->maybe_apply(f, x, p, t), (; kwargs...))...)
    simfun(x, p, t) = func(x, p, t; map(f->maybe_apply(f, x, p, t), (; kwargs...))...)
    return simfun
end

softsign(x) = tanh(1e3x)
```

--------------------------------

### Combine and Plot Responses

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Combine the step and sine response plots into a single figure for comparison. Assumes `step_p` and `sin_p` are defined.

```julia
plot(step_p, sin_p; layout = (2, 1), size = (800, 800))
```

--------------------------------

### Calculate equivalent viscous damping

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/coulomb_control

Calculates the equivalent viscous damping coefficient for a system, based on Coulomb friction parameters. This is useful for comparing different friction models.

```julia
# Equivalent viscous damping coefficient taken from:
# https://engineering.purdue.edu/~deadams/ME563/lecture2010.pdf
visc_equiv(μ, N, ω, mag) = 4*μ*N/(π*ω*mag)
```

--------------------------------

### KeepIndex with Range Index

Source: https://docs.sciml.ai/ComponentArrays/stable/indexing_behavior

Illustrates using `KeepIndex` with a range to capture multiple elements and retain the names of any full components captured within the range.

```julia
ca[KeepIndex(2:3)]
```

--------------------------------

### Define Lotka-Volterra System ODE and Jacobian

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/ODE_jac

Defines the ODE function `lotka!` and its Jacobian function `lotka_jac!` for the Lotka-Volterra system using ComponentArrays. The Jacobian function uses symbolic indexing.

```julia
## Lotka-Volterra system
function lotka!(D, u, p, t; f = 0.0)
    @unpack α, β, γ, δ = p
    @unpack x, y = u

    D.x = α*x - β*x*y + f
    D.y = -γ*y + δ*x*y
    return nothing
end
function lotka_jac!(D, u, p, t)
    @unpack α, β, γ, δ = p
    @unpack x, y = u

    D[:x, :x] = α - β*y
    D[:x, :y] = -β*x

    D[:y, :x] = δ*y
    D[:y, :y] = -γ + δ*x
    return nothing
end

lotka_p = (α = 2/3, β = 4/3, γ = 1.0, δ = 1.0)
lotka_ic = ComponentArray(x = 1.0, y = 1.0)
lotka_fun = ODEFunction(lotka!, jac = lotka_jac!)
lotka_prob = ODEProblem(lotka_fun, lotka_ic, tspan, lotka_p)
```

--------------------------------

### Define Laplace Variable

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Initialize the Laplace variable 's' for use in transfer function definitions.

```julia
s = tf("s")
```

--------------------------------

### Access and Modify ComponentArray Properties using Dot Notation

Source: https://docs.sciml.ai/ComponentArrays/stable/indexing_behavior

Demonstrates accessing and modifying components of a ComponentArray using dot notation, similar to structs or NamedTuples.

```julia
ca.b[1] = 99
ca.a = 22
ca
```

--------------------------------

### Define Lorenz System ODE and Jacobian

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/ODE_jac

Defines the ODE function `lorenz!` and its Jacobian function `lorenz_jac!` for the Lorenz system using ComponentArrays. The Jacobian function uses symbolic indexing for clarity.

```julia
using ComponentArrays
using DifferentialEquations
using Parameters: @unpack

tspan = (0.0, 20.0)

## Lorenz system
function lorenz!(D, u, p, t; f = 0.0)
    @unpack σ, ρ, β = p
    @unpack x, y, z = u

    D.x = σ*(y - x)
    D.y = x*(ρ - z) - y - f
    D.z = x*y - β*z
    return nothing
end
function lorenz_jac!(D, u, p, t)
    @unpack σ, ρ, β = p
    @unpack x, y, z = u

    D[:x, :x] = -σ
    D[:x, :y] = σ

    D[:y, :x] = ρ
    D[:y, :y] = -1
    D[:y, :z] = -x

    D[:z, :x] = y
    D[:z, :y] = x
    D[:z, :z] = -β
    return nothing
end

lorenz_p = (σ = 10.0, ρ = 28.0, β = 8/3)
lorenz_ic = ComponentArray(x = 0.0, y = 0.0, z = 0.0)
lorenz_fun = ODEFunction(lorenz!, jac = lorenz_jac!)
lorenz_prob = ODEProblem(lorenz_fun, lorenz_ic, tspan, lorenz_p)
```

--------------------------------

### Retain Component Labels using KeepIndex Wrapper

Source: https://docs.sciml.ai/ComponentArrays/stable/indexing_behavior

Demonstrates using the `KeepIndex` wrapper to index into a ComponentArray and retain the component name, returning a ComponentVector instead of a plain array.

```julia
ca[KeepIndex(:b)]
```

--------------------------------

### Create a Nested ComponentArray

Source: https://docs.sciml.ai/ComponentArrays/stable/indexing_behavior

Initializes a ComponentArray with a deeply nested structure, including named scalar and vector components within a nested tuple.

```julia
ca = ComponentArray(a = 5, b = [4, 1], c = (a = 2, b = [6, 30]))
```

--------------------------------

### Control Law Definition

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/adaptive_control

Defines the basic control law equation.

```julia
control(θ, w) = θ'w
```

--------------------------------

### ComponentArrays.valkeys

Source: https://docs.sciml.ai/ComponentArrays/stable/api

Returns `Val`-wrapped keys of `ComponentVector` for fast iteration over component keys. Also works directly on an `AbstractAxis`.

```APIDOC
## valkeys(x)

### Description
Returns `Val`-wrapped keys of `ComponentVector` for fast iteration over component keys. Also works directly on an `AbstractAxis`.

### Parameters
#### Path Parameters
- **x** (ComponentVector or AbstractAxis) - The ComponentVector or AbstractAxis to get keys from.

### Response
#### Success Response
- Returns an iterable collection of `Val`-wrapped keys.
```

--------------------------------

### Prediction and Loss Function

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/DiffEqFlux

Defines functions to predict ODE solutions using the neural network and to calculate the loss based on the difference from true data.

```julia
predict_n_ode(θ) = Array(solve(prob, Tsit5(), u0 = θ.u, p = θ.p, saveat = t))

function loss_n_ode(θ)
    pred = predict_n_ode(θ)
    loss = sum(abs2, ode_data .- pred)
    return loss, pred
end
loss_n_ode(θ)
```

--------------------------------

### Define Composed System ODE and Jacobian

Source: https://docs.sciml.ai/ComponentArrays/stable/examples/ODE_jac

Defines the ODE function `composed!` and its Jacobian function `composed_jac!` for a system composed of Lorenz and Lotka-Volterra models using ComponentArrays. It demonstrates how to handle nested ComponentArrays and partial derivatives.

```julia
## Composed Lorenz and Lotka-Volterra system
function composed!(D, u, p, t)
    c = p.c #coupling parameter
    @unpack lorenz, lotka = u

    lorenz!(D.lorenz, lorenz, p.lorenz, t, f = c*lotka.x)
    lotka!(D.lotka, lotka, p.lotka, t, f = c*lorenz.x)
    return nothing
end
function composed_jac!(D, u, p, t)
    c = p.c
    @unpack lorenz, lotka = u

    lorenz_jac!(@view(D[:lorenz, :lorenz]), lorenz, p.lorenz, t)
    lotka_jac!(@view(D[:lotka, :lotka]), lotka, p.lotka, t)

    @view(D[:lorenz, :lotka])[:y, :x] = -c
    @view(D[:lotka, :lorenz])[:x, :x] = c
    return nothing
end

comp_p = (lorenz = lorenz_p, lotka = lotka_p, c = 0.01)
comp_ic = ComponentArray(lorenz = lorenz_ic, lotka = lotka_ic)
comp_fun = ODEFunction(composed!, jac = composed_jac!)
comp_prob = ODEProblem(comp_fun, comp_ic, tspan, comp_p)
```

--------------------------------

### KeepIndex with Range to End

Source: https://docs.sciml.ai/ComponentArrays/stable/indexing_behavior

Shows how `KeepIndex` with a range to the end captures elements and retains the names of the components fully included in the range.

```julia
ca[KeepIndex(2:end)]
```

--------------------------------

### ComponentArrays.ViewAxis

Source: https://docs.sciml.ai/ComponentArrays/stable/api

An Axis used for creating arrays of ComponentArrays, specifying a parent index and an index map.

```APIDOC
## `ComponentArrays.ViewAxis` — Type

### Description

Axis for creating arrays of `ComponentArray`s.

### Usage

```julia
va = ViewAxis(parent_index, index_map)
```
```