systems.base.utils.ObservationEngine

systems.base.utils.ObservationEngine(system, code_gen, backend_mgr)

Evaluates observation/output functions across backends.

Handles the evaluation of y = h(x) and C = ∂h/∂x for all backends with proper shape handling and batching.

Type System Integration: - StateVector: Input state - OutputVector: Evaluated output y = h(x) - OutputMatrix: Linearized observation C = ∂h/∂x - ObservationLinearization: (C, D) tuple for full output linearization - Backend: Type-safe backend selection

Batching: Supports both single and batched evaluation using centralized utilities from the type framework (is_batched, get_batch_size).

Example: >>> engine = ObservationEngine(system, code_gen, backend_mgr) >>> y: OutputVector = engine.evaluate(x, backend=‘numpy’) >>> C: OutputMatrix = engine.compute_jacobian(x, backend=‘numpy’)

Methods

Name	Description
compute_jacobian	Compute linearized observation: C = ∂h/∂x.
compute_symbolic	Compute symbolic linearization C = ∂h/∂x.
evaluate	Evaluate output equation: y = h(x).

compute_jacobian

systems.base.utils.ObservationEngine.compute_jacobian(x, backend=None)

Compute linearized observation: C = ∂h/∂x.

If no custom output function, returns identity matrix.

Args: x: State at which to linearize backend: Backend selection: - None: Auto-detect from input type (default) - ‘numpy’, ‘torch’, ‘jax’: Force specific backend - ‘default’: Use system’s default backend

Returns: OutputMatrix C matrix (ny, nx) - output Jacobian (type matches backend)

Example: >>> C: OutputMatrix = engine.compute_jacobian(x, backend=‘numpy’) >>> print(C.shape) # (ny, nx)

compute_symbolic

systems.base.utils.ObservationEngine.compute_symbolic(x_eq=None)

Compute symbolic linearization C = ∂h/∂x.

Args: x_eq: Equilibrium state (zeros if None)

Returns: C: Symbolic Jacobian matrix

Example: >>> C_sym = engine.compute_symbolic(x_eq=sp.Matrix([0, 0]))

evaluate

systems.base.utils.ObservationEngine.evaluate(x, backend=None)

Evaluate output equation: y = h(x).

If no custom output function is defined, returns the full state (identity).

Args: x: State vector backend: Backend selection: - None: Auto-detect from input type (default) - ‘numpy’, ‘torch’, ‘jax’: Force specific backend - ‘default’: Use system’s default backend

Returns: Output vector (type matches backend)

Example: >>> y: OutputVector = engine.evaluate(x_numpy) # Auto-detect NumPy >>> y: OutputVector = engine.evaluate(x_numpy, backend=‘torch’) # Convert to torch

# systems.base.utils.ObservationEngine { #cdesym.systems.base.utils.ObservationEngine } ```python systems.base.utils.ObservationEngine(system, code_gen, backend_mgr) ``` Evaluates observation/output functions across backends. Handles the evaluation of y = h(x) and C = ∂h/∂x for all backends with proper shape handling and batching. Type System Integration: - StateVector: Input state - OutputVector: Evaluated output y = h(x) - OutputMatrix: Linearized observation C = ∂h/∂x - ObservationLinearization: (C, D) tuple for full output linearization - Backend: Type-safe backend selection Batching: Supports both single and batched evaluation using centralized utilities from the type framework (is_batched, get_batch_size). Example: >>> engine = ObservationEngine(system, code_gen, backend_mgr) >>> y: OutputVector = engine.evaluate(x, backend='numpy') >>> C: OutputMatrix = engine.compute_jacobian(x, backend='numpy') ## Methods | Name | Description | | --- | --- | | [compute_jacobian](#cdesym.systems.base.utils.ObservationEngine.compute_jacobian) | Compute linearized observation: C = ∂h/∂x. | | [compute_symbolic](#cdesym.systems.base.utils.ObservationEngine.compute_symbolic) | Compute symbolic linearization C = ∂h/∂x. | | [evaluate](#cdesym.systems.base.utils.ObservationEngine.evaluate) | Evaluate output equation: y = h(x). | ### compute_jacobian { #cdesym.systems.base.utils.ObservationEngine.compute_jacobian } ```python systems.base.utils.ObservationEngine.compute_jacobian(x, backend=None) ``` Compute linearized observation: C = ∂h/∂x. If no custom output function, returns identity matrix. Args: x: State at which to linearize backend: Backend selection: - None: Auto-detect from input type (default) - 'numpy', 'torch', 'jax': Force specific backend - 'default': Use system's default backend Returns: OutputMatrix C matrix (ny, nx) - output Jacobian (type matches backend) Example: >>> C: OutputMatrix = engine.compute_jacobian(x, backend='numpy') >>> print(C.shape) # (ny, nx) ### compute_symbolic { #cdesym.systems.base.utils.ObservationEngine.compute_symbolic } ```python systems.base.utils.ObservationEngine.compute_symbolic(x_eq=None) ``` Compute symbolic linearization C = ∂h/∂x. Args: x_eq: Equilibrium state (zeros if None) Returns: C: Symbolic Jacobian matrix Example: >>> C_sym = engine.compute_symbolic(x_eq=sp.Matrix([0, 0])) ### evaluate { #cdesym.systems.base.utils.ObservationEngine.evaluate } ```python systems.base.utils.ObservationEngine.evaluate(x, backend=None) ``` Evaluate output equation: y = h(x). If no custom output function is defined, returns the full state (identity). Args: x: State vector backend: Backend selection: - None: Auto-detect from input type (default) - 'numpy', 'torch', 'jax': Force specific backend - 'default': Use system's default backend Returns: Output vector (type matches backend) Example: >>> y: OutputVector = engine.evaluate(x_numpy) # Auto-detect NumPy >>> y: OutputVector = engine.evaluate(x_numpy, backend='torch') # Convert to torch