Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

added k-means initialisation for Poisson HMMs #358

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

added k-means initialisation for Poisson HMMs #358

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5d67795
added k-means initialisation for Poisson HMMs
anuj-sp Feb 15, 2024
359d86e
fixed k-means initialization in PoissonHMM class
anuj-sp Feb 15, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 12 additions & 4 deletions dynamax/hidden_markov_model/models/poisson_hmm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -45,14 +45,21 @@ def emission_shape(self):

def initialize(self, key=jr.PRNGKey(0),
method="prior",
emission_rates=None):
emission_rates=None,
emissions=None):
# Initialize the emission probabilities
if emission_rates is None:
if method.lower() == "prior":
prior = tfd.Gamma(self.emission_prior_concentration, self.emission_prior_rate)
emission_rates = prior.sample(seed=key, sample_shape=(self.num_states, self.emission_dim))
elif method.lower() == "kmeans":
raise NotImplementedError("kmeans initialization is not yet implemented!")
assert emissions is not None, "Need emissions to initialize the model with K-Means!"
from sklearn.cluster import KMeans
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(emissions.reshape(-1, self.emission_dim))
## Cluster centers, also forms the Poisson emission rate
emission_rates = jnp.array(km.cluster_centers_)
else:
raise Exception("invalid initialization method: {}".format(method))
else:
Expand Down Expand Up @@ -136,7 +143,8 @@ def initialize(self, key=jr.PRNGKey(0),
method="prior",
initial_probs: Optional[Float[Array, "num_states"]]=None,
transition_matrix: Optional[Float[Array, "num_states num_states"]]=None,
emission_rates: Optional[Float[Array, "num_states emission_dim"]]=None
emission_rates: Optional[Float[Array, "num_states emission_dim"]]=None,
emissions: Optional[Float[Array, "num_timesteps emission_dim"]]=None
) -> Tuple[ParameterSet, PropertySet]:
"""Initialize the model parameters and their corresponding properties.

Expand All @@ -160,5 +168,5 @@ def initialize(self, key=jr.PRNGKey(0),
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_rates=emission_rates)
params["emissions"], props["emissions"] = self.emission_component.initialize(key3, method=method, emission_rates=emission_rates, emissions = emissions)
return ParamsPoissonHMM(**params), ParamsPoissonHMM(**props)