dpmmlearn is a algorithms for Dirichlet Process Mixture Model.
The required dependencies to use dpmmlearn are,
- scikit-learn
- numpy
- scipy
You also need matplotlib, seaborn to run the demo and pytest to run the tests.
pip install dpmmlearnWe have posted a usage example in the demo folder.
from dpmmlearn import DPMM
from dpmmlearn.probability import NormInvWish
mu_0 = 0.0
kappa_0 = 1.0
Lam_0 = np.eye(2) * 10
nu_0 = 2
alpha = 3.0
prob = NormInvWish(mu_0, kappa_0, Lam_0, nu_0)
model = DPMM(prob, alpha, max_iter=300, random_state=0, verbose=True)
model.fit(X)
labels = model.labels_from dpmmlearn import DPMM
from dpmmlearn.probability import NormInvChi2
def sample_pdf(x, mu_list, var_list, portion):
y = np.zeros_like(x)
for mu, var, num in zip(mu_list, var_list, portion):
y_ = np.exp(-0.5*(x-mu)**2/var)/np.sqrt(2*np.pi*var)
y += num/sum(portion) * y_
return y
mu_0 = 0.3
kappa_0 = 0.1
sigsqr_0 = 0.1
nu_0 = 1.0
alpha = 1.0
prob = NormInvChi2(mu_0, kappa_0, sigsqr_0, nu_0)
model = DPMM(prob, alpha, max_iter=500, random_state=0, verbose=True)
model.fit(X)
mu_ = [mu for mu, var in model.thetas_]
var_ = [var for mu, var in model.thetas_]
portion_ = [k/sum(model.n_labels_) for k in model.n_labels_]
x = np.arange(-2, 2, 0.01)
y_pred = sample_pdf(x, mu_, var_, portion_)from dpmmlearn import DPMM
from dpmmlearn.probability import InvGamma
def result_pdf(x, thetas, n_labels, mu):
y = np.zeros_like(x)
for theta, n_label in zip(thetas, n_labels):
var = theta
y_ = np.exp(-0.5*(x-mu)**2/var)/np.sqrt(2*np.pi*var)
y += n_label/sum(n_labels) * y_
return y
mu = 0.0
alpha = 1.0
beta = 100.0
dp_alpha = 0.1
prob = InvGamma(alpha, beta, mu)
model = DPMM(prob, dp_alpha, max_iter=500, random_state=0, verbose=True)
model.fit(X)
y_pred = result_pdf(x, model.thetas_, model.n_labels_, mu)This code is licensed under MIT License.
python setup.py test