Configuration¶
The following are configuration options for various Sequentia classes and functions.
API Reference¶
Covariance matrix types for |
|
Prior probability types for |
|
Topology types for Hidden Markov Models. |
|
Initial state and transition probability types for Hidden Markov Models. |
Configuration values for Sequentia classes and functions.
- enum sequentia.enums.CovarianceMode(value)¶
Covariance matrix types for
GaussianMixtureHMM
.- Member Type:
Valid values are as follows:
- FULL = <CovarianceMode.FULL: 'full'>¶
All values are fully learnable independently for each component.
- DIAGONAL = <CovarianceMode.DIAGONAL: 'diag'>¶
Only values along the diagonal may be learned independently for each component.
- enum sequentia.enums.PriorMode(value)¶
Prior probability types for
HMMClassifier
.- Member Type:
Valid values are as follows:
- UNIFORM = <PriorMode.UNIFORM: 'uniform'>¶
Equal probability for each class.
- FREQUENCY = <PriorMode.FREQUENCY: 'frequency'>¶
Inverse count of the occurrences of the class in the training data.
- enum sequentia.enums.TopologyMode(value)¶
Topology types for Hidden Markov Models.
- Member Type:
Valid values are as follows:
- ERGODIC = <TopologyMode.ERGODIC: 'ergodic'>¶
All states have a non-zero probability of transitioning to any state.
- LEFT_RIGHT = <TopologyMode.LEFT_RIGHT: 'left-right'>¶
States are arranged in a way such that any state may only transition to itself or any state ahead of it, but not to any previous state.
- LINEAR = <TopologyMode.LINEAR: 'linear'>¶
Same as
LEFT_RIGHT
, but states are only permitted to transition to the next state.
- enum sequentia.enums.TransitionMode(value)¶
Initial state and transition probability types for Hidden Markov Models.
- Member Type:
Valid values are as follows:
- UNIFORM = <TransitionMode.UNIFORM: 'uniform'>¶
Equal probability of starting in or transitioning to each state according to the topology.
- RANDOM = <TransitionMode.RANDOM: 'random'>¶
Random probability of starting in or transitioning to each state according to the topology. State probabilities are sampled from a Dirichlet distribution with unit concentration parameters.