pymc-devs/pymc-extras

AI Summary: This issue proposes replacing Pandas with Narwhals in the `PyMCStatespace` class and its associated data preprocessing utilities. The goal is to make the data handling more agnostic, allowing users to provide data from various dataframe libraries, as the current implementation is tightly coupled to Pandas, even depending on index types. Narwhals is suggested as a suitable abstraction layer to achieve this flexibility.

AI Summary: Refactor the `test_structural_model_against_statsmodels` test in the `pymc-extras` project. The task involves splitting the existing test into smaller, more focused tests, each verifying a specific property of the structural model components. This improves maintainability and debuggability by reducing complexity and eliminating redundant checks.

AI Summary: The task is to add a check to the `build_statespace_graph` function in the `pymc-extras` project to ensure that the order of columns in the input dataframe matches the order of observed states specified in `observed_state_names`. This is crucial for multivariate time series to prevent misalignment of time series indices, ensuring correct model behavior.

AI Summary: The task involves creating a custom C operator (COp) for Kalman filtering within the PyMC ecosystem. This aims to improve performance by bypassing the limitations of JAX and enabling support for non-standard operations and sparse matrices, potentially as part of the `pymc-extras` repository for experimental features.

AI Summary: Explore the feasibility and implementation of using sparse matrices within the PyMC statespace models to handle large, sparse systems. This involves investigating PyTensor's sparse matrix capabilities, determining necessary modifications to the KalmanFilter and related components (PytensorRepresentation, PyMCStatespace, model.build_symbolic_graph methods) to support sparse inputs and outputs.