import jax.numpy as jnp
import jax.random as jr
import tensorflow_probability.substrates.jax.distributions as tfd
import tensorflow_probability.substrates.jax.bijectors as tfb
from jaxtyping import Float, Array
from dynamax.parameters import ParameterProperties
from dynamax.hidden_markov_model.models.abstractions import HMM, HMMEmissions, HMMParameterSet, HMMPropertySet
from dynamax.hidden_markov_model.models.initial import StandardHMMInitialState, ParamsStandardHMMInitialState
from dynamax.hidden_markov_model.models.transitions import StandardHMMTransitions, ParamsStandardHMMTransitions
from dynamax.types import Scalar
import optax
from typing import NamedTuple, Optional, Tuple, Union
class ParamsLogisticRegressionHMMEmissions(NamedTuple):
weights: Union[Float[Array, "state_dim input_dim"], ParameterProperties]
biases: Union[Float[Array, "state_dim"], ParameterProperties]
class LogisticRegressionHMMEmissions(HMMEmissions):
def __init__(self,
num_states,
input_dim,
emission_matrices_scale=1e8,
m_step_optimizer=optax.adam(1e-2),
m_step_num_iters=50):
super().__init__(m_step_optimizer=m_step_optimizer, m_step_num_iters=m_step_num_iters)
self.num_states = num_states
self.input_dim = input_dim
self.emission_weights_scale = emission_matrices_scale
@property
def emission_shape(self):
return ()
@property
def inputs_shape(self):
return (self.input_dim,)
def initialize(self,
key=jr.PRNGKey(0),
method="prior",
emission_weights=None,
emission_biases=None,
emissions=None,
inputs=None):
if method.lower() == "kmeans":
assert emissions is not None, "Need emissions to initialize the model with K-Means!"
assert inputs is not None, "Need inputs to initialize the model with K-Means!"
from sklearn.cluster import KMeans
flat_emissions = emissions.reshape(-1,)
flat_inputs = inputs.reshape(-1, self.input_dim)
key, subkey = jr.split(key) # Create a random seed for SKLearn.
sklearn_key = jr.randint(subkey, shape=(), minval=0, maxval=2147483647) # Max int32 value.
km = KMeans(self.num_states, random_state=int(sklearn_key)).fit(flat_inputs)
_emission_weights = jnp.zeros((self.num_states, self.input_dim))
_emission_biases = jnp.array([tfb.Sigmoid().inverse(flat_emissions[km.labels_ == k].mean())
for k in range(self.num_states)])
elif method.lower() == "prior":
# TODO: Use an MNIW prior
key1, key2, key = jr.split(key, 3)
_emission_weights = 0.01 * jr.normal(key1, (self.num_states, self.input_dim))
_emission_biases = jr.normal(key2, (self.num_states,))
else:
raise Exception("Invalid initialization method: {}".format(method))
# Only use the values above if the user hasn't specified their own
default = lambda x, x0: x if x is not None else x0
params = ParamsLogisticRegressionHMMEmissions(
weights=default(emission_weights, _emission_weights),
biases=default(emission_biases, _emission_biases))
props = ParamsLogisticRegressionHMMEmissions(
weights=ParameterProperties(),
biases=ParameterProperties())
return params, props
def log_prior(self, params):
return tfd.Normal(0, self.emission_weights_scale).log_prob(params.weights).sum()
def distribution(self, params, state, inputs):
logits = params.weights[state] @ inputs + params.biases[state]
return tfd.Bernoulli(logits=logits)
class ParamsLogisticRegressionHMM(NamedTuple):
initial: ParamsStandardHMMInitialState
transitions: ParamsStandardHMMTransitions
emissions: ParamsLogisticRegressionHMMEmissions
[docs]
class LogisticRegressionHMM(HMM):
r"""An HMM whose emissions come from a logistic regression with state-dependent weights.
This is also known as a *switching logistic regression* model.
Let $y_t \in \{0,1\}$ and $u_t \in \mathbb{R}^M$ denote binary emissions
and inputs at time $t$, respectively. In this model, the emission distribution is,
$$p(y_t \mid z_t, u_t, \theta) = \mathrm{Bern}(y_{t} \mid \sigma(w_{z_t}^\top u_t + b_{z_t}))$$
with *emission weights* $w_k \in \mathbb{R}^{M}$ and *emission biases* $b_k \in \mathbb{R}$.
We use $L_2$ regularization on the emission weights, which can be thought of as a
Gaussian prior,
$$p(\theta) \propto \prod_{k=1}^K \prod_{m=1}^M \mathcal{N}(w_{k,m} \mid 0, \varsigma^2)$$
:param num_states: number of discrete states $K$
:param input_dim: input dimension $M$
:param initial_probs_concentration: $\alpha$
:param transition_matrix_concentration: $\beta$
:param transition_matrix_stickiness: optional hyperparameter to boost the concentration on the diagonal of the transition matrix.
:param emission_matrices_scale: $\varsigma$
:param m_step_optimizer: ``optax`` optimizer, like Adam.
:param m_step_num_iters: number of optimizer steps per M-step.
"""
def __init__(self,
num_states: int,
input_dim: int,
initial_probs_concentration: Union[Scalar, Float[Array, "num_states"]]=1.1,
transition_matrix_concentration: Union[Scalar, Float[Array, "num_states"]]=1.1,
transition_matrix_stickiness: Scalar=0.0,
emission_matrices_scale: Scalar=1e8,
m_step_optimizer: optax.GradientTransformation=optax.adam(1e-2),
m_step_num_iters: int=50):
self.inputs_dim = input_dim
initial_component = StandardHMMInitialState(num_states, initial_probs_concentration=initial_probs_concentration)
transition_component = StandardHMMTransitions(num_states, concentration=transition_matrix_concentration, stickiness=transition_matrix_stickiness)
emission_component = LogisticRegressionHMMEmissions(num_states, input_dim, emission_matrices_scale=emission_matrices_scale, m_step_optimizer=m_step_optimizer, m_step_num_iters=m_step_num_iters)
super().__init__(num_states, initial_component, transition_component, emission_component)
@property
def inputs_shape(self):
return (self.inputs_dim,)
[docs]
def initialize(self,
key: jr.PRNGKey=jr.PRNGKey(0),
method: str="prior",
initial_probs: Optional[Float[Array, "num_states"]]=None,
transition_matrix: Optional[Float[Array, "num_states num_states"]]=None,
emission_weights: Optional[Float[Array, "num_states input_dim"]]=None,
emission_biases: Optional[Float[Array, "num_states"]]=None,
emissions: Optional[Float[Array, "num_timesteps"]]=None,
inputs: Optional[Float[Array, "num_timesteps input_dim"]]=None,
) -> Tuple[HMMParameterSet, HMMPropertySet]:
"""Initialize the model parameters and their corresponding properties.
You can either specify parameters manually via the keyword arguments, or you can have
them set automatically. If any parameters are not specified, you must supply a PRNGKey.
Parameters will then be sampled from the prior (if `method==prior`).
Args:
key: random number generator for unspecified parameters. Must not be None if there are any unspecified parameters.
method: method for initializing unspecified parameters. Both "prior" and "kmeans" are supported.
initial_probs: manually specified initial state probabilities.
transition_matrix: manually specified transition matrix.
emission_weights: manually specified emission weights.
emission_biases: manually specified emission biases.
emissions: emissions for initializing the parameters with kmeans.
inputs: inputs for initializing the parameters with kmeans.
Returns:
Model parameters and their properties.
"""
key1, key2, key3 = jr.split(key , 3)
params, props = dict(), dict()
params["initial"], props["initial"] = self.initial_component.initialize(key1, method=method, initial_probs=initial_probs)
params["transitions"], props["transitions"] = self.transition_component.initialize(key2, method=method, transition_matrix=transition_matrix)
params["emissions"], props["emissions"] = self.emission_component.initialize(key3, method=method, emission_weights=emission_weights, emission_biases=emission_biases, emissions=emissions, inputs=inputs)
return ParamsLogisticRegressionHMM(**params), ParamsLogisticRegressionHMM(**props)
