liveBook · Manning

State space model

As explained in Grokking Bayes

A model that represents a system using an unobserved state that evolves over time and an observation model connecting that state to measured data.

A state-space model describes a system whose true state is hidden but changes over time. It separates the evolving latent state from the observations generated by that state. section 9.2

This separation fits time series such as atmospheric CO₂: the underlying concentration evolves gradually, while measurements include noise. The transition model connects future states to current states; the observation model connects states to measured data. section 9.2

Sequential inference follows a repeating cycle:

  • Specify how the latent state evolves with a state transition model. section 9.3
  • Predict the next state using the current filtered belief and transition model. section 9.3
  • Update that prediction when a new observation arrives through the observation model. section 9.3
  • Repeat the predict-and-update cycle to track beliefs, uncertainty, and forecasts over time. section 9.0

Atmospheric CO₂

Treat the true CO₂ concentration in each year as the hidden state and the recorded concentration as a noisy observation. A filter predicts the next year's concentration, then corrects it using the new measurement. section 9.2 section 9.3

Figure 9.3 A diagram of a state space model, which includes an unknown latent space 𝑥ᵢ that evolves over time, and the observation model, which describes the relationship between the latent state and observed data 𝑦ᵢ at each time step.

Use the diagram to distinguish the hidden state sequence from the observed data sequence and their relationship at each time step. section 9.2

How Grokking Bayes covers it

How deep it goes
Defines latent states, transitions, observations, and recursive filtering, then stops short of detailed implementation equations.
Examples
Uses atmospheric CO₂ measurements to illustrate hidden concentration, noisy observations, temporal dependence, prediction, and updating.
What you can do after
Apply it to formulate sequential time-series inference by separating latent dynamics from observations and alternating prediction with updating.
How it connects
Connects state transition models, observation models, filters, Bayesian inference, time series, and Kalman filters.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Grokking Bayes. The section links point to the parts of the book it draws on.