pytorch_hamiliton_unittest

What problem does it solve?

It prevents subtle mathematical and tensor-shape regressions in vectorized PyTorch Markov Regime-Switching models by enforcing domain constraints and probabilistic invariants during automated unit testing.

Core Features & Use Cases

• Reparameterization validation: Ensures raw parameters map into safe domains, including strictly positive standard deviations and properly normalized, non-negative transition probabilities.
• Hamilton filter invariants: Verifies that filtered state assignment probabilities are well-formed at every time step (non-negative and summing to 1.0).
• Regime identifiability checks: Confirms permutation equivariance when sorting regime states, including consistent reordering across means, sigma parameters, and multi-dimensional transition tensors.
• Input validation: Asserts expected tensor dimensions and raises errors for mismatched inputs, protecting downstream training and inference.

Quick Start

Use the pytorch_hamiliton_unittest skill to run the unit tests in examples/test_mrs_model.py and verify constraint, sorting, forward-pass, and prediction-step behaviors for your Markov Regime-Switching implementation.