markov-regime-features

Markov regime features remain uniform when using default priors without stateful tracking. Use InferenceObservationBuilder with window=100 and use_gpu_markov=True to maintain Markov state across price history, allowing vol_prob and trend_prob to evolve dynamically instead of staying frozen.

Uniform probabilities indicate stale or missing Markov state. The Markov regime inference requires stateful tracking that updates vol_prob_low/med/high and trend_prob_down/neutral/up based on incoming price data; without it, priors never change.

Create an InferenceObservationBuilder, configure it with window=100 and use_gpu_markov=True for GPU acceleration, then call build() with prices, high, and low arrays. The builder maintains state across calls and outputs evolving vol_probs and trend_probs reflecting current regime estimates.

Yes. After building observations with InferenceObservationBuilder, inspect vol_probs and trend_probs across your observation sequence to see regime evolution. This reveals whether probabilities adapt to price volatility and trend changes or remain stationary.

GPU acceleration is optional but recommended for performance. Set use_gpu_markov=True in InferenceObservationBuilder to enable it; this accelerates stateful inference across large price histories, though CPU execution is supported.

Window=100 is the recommended starting point for InferenceObservationBuilder; it balances Markov state freshness with sufficient historical context to estimate volatility and trend regimes from price movements.