methylseg.StickyCategoricalMethylSegHMM

class methylseg.StickyCategoricalMethylSegHMM(n_states, random_state=42, n_iter=30, stay_prob=0.99995, emission_mismatch_prob=0.45, transition_prior_strength=50.0, fit_transitions=False)[source]

Bases: MethylSegHMM

Categorical HMM with strong self-transition priors for smoother segments.

Parameters:

n_states (int)
random_state (int)
n_iter (int)
stay_prob (float)
emission_mismatch_prob (float)
transition_prior_strength (float)
fit_transitions (bool)

__init__(n_states, random_state=42, n_iter=30, stay_prob=0.99995, emission_mismatch_prob=0.45, transition_prior_strength=50.0, fit_transitions=False)[source]

Initialize a sticky categorical HMM.

Parameters:

n_states (int) – Number of hidden and observed categorical states.
random_state (int) – Random seed passed to the underlying HMM implementation.
n_iter (int) – Maximum number of expectation-maximization iterations.
stay_prob (float) – Initial self-transition probability for each hidden state.
emission_mismatch_prob (float) – Probability mass assigned to nonmatching observed states.
transition_prior_strength (float) – Weight of the sticky transition prior when transitions are fitted.
fit_transitions (bool) – If True, estimate transitions during fitting; otherwise retain the initialized sticky transition matrix.

Methods

`__init__`(n_states[, random_state, n_iter, ...])	Initialize a sticky categorical HMM.
`create_model`()	Create and initialize the sticky categorical HMM backend.
`fit`(emissions[, sample_info, chrom])	Fit transition probabilities when `fit_transitions` is enabled.
`format_fit`(emissions)	Reshape integer categorical observations for `CategoricalHMM`.
`format_predict`(emissions)	Format categorical observations for prediction.
`make_emissionprob`()	Build the near-identity emission matrix used by the sticky HMM.
`make_sticky_transmat`([n_states, stay_prob])	Build a 'sticky' transition matrix.
`predict`(emissions)	Predict smoothed hidden states from categorical observations.

Attributes

`DEFAULT_STAY_PROB`
`EMISSION_MISMATCH_PROB`
`TRANSITION_PRIOR_STRENGTH`

DEFAULT_STAY_PROB = 0.99995

EMISSION_MISMATCH_PROB = 0.45

TRANSITION_PRIOR_STRENGTH = 50.0

format_fit(emissions)[source]

Reshape integer categorical observations for CategoricalHMM.

Parameters:: emissions – Integer observation labels in [0, n_states).
Returns:: Column vector with one categorical code per observation.
Return type:: numpy.ndarray

format_predict(emissions)[source]

Format categorical observations for prediction.

Parameters:: emissions – Integer observation labels in [0, n_states).
Returns:: Column vector with one categorical code per observation.
Return type:: numpy.ndarray

make_sticky_transmat(n_states=None, stay_prob=None)[source]

Build a ‘sticky’ transition matrix.

Parameters:

n_states (int | None) – Optional number of hidden states. When omitted, uses self.n_states.
stay_prob (float | None) – Optional diagonal self-transition probability. When omitted, uses self.stay_prob.

Returns:

numpy.ndarray – Square transition matrix whose rows sum to one.
stay_prob controls the diagonal self-transition probability for every
state. Remaining mass is shared uniformly across off-diagonal entries.

Return type:

ndarray

make_emissionprob()[source]

Build the near-identity emission matrix used by the sticky HMM.

Returns:: Square emission-probability matrix whose diagonal retains most of the probability mass.
Return type:: numpy.ndarray

create_model()[source]: Create and initialize the sticky categorical HMM backend.

fit(emissions, sample_info=None, chrom=None)[source]

Fit transition probabilities when fit_transitions is enabled.

Parameters:

emissions – Integer categorical observations.
sample_info – Unused placeholder for API compatibility.
chrom – Unused placeholder for API compatibility.

predict(emissions)[source]

Predict smoothed hidden states from categorical observations.

Parameters:: emissions – Integer categorical observations.
Returns:: Decoded HMM state sequence.
Return type:: numpy.ndarray