def gaussian_exhaustive_test():
model = random_model(3)
obs_distns = [
d.Gaussian(mu=np.random.randn(2),sigma=np.eye(2)),
d.Gaussian(mu=np.random.randn(2),sigma=np.eye(2)),
d.Gaussian(mu=np.random.randn(2),sigma=np.eye(2))]
stateseq = np.random.randint(3,size=10)
data = np.vstack([obs_distns[a].rvs() for a in stateseq])
target_val = compute_likelihood_enumeration(obs_distns=obs_distns,data=data,**model)
likelihood_check(target_val=target_val,data=data,obs_distns=obs_distns,**model)
def _random_variant_model():
N = 4
obs_dim = 2
obs_hypparams = {
'mu_0': np.zeros(obs_dim),
'sigma_0': np.eye(obs_dim),
'kappa_0': 0.05,
'nu_0': obs_dim + 5
}
obs_distns = [d.Gaussian(**obs_hypparams) for state in range(N)]
dur_distns = \
[d.NegativeBinomialIntegerRVariantDuration(
np.r_[0,0,0,0,0,1.,1.,1.], # discrete distribution uniform over {6,7,8}
alpha_0=9,beta_0=1, # average geometric success probability 1/(9+1)
) for state in range(N)]
model = m.HSMMIntNegBinVariant(init_state_concentration=10.,
alpha=6.,
gamma=6.,
obs_distns=obs_distns,
dur_distns=dur_distns)
return model
def likelihood_exhaustive_tests():
### discrete data
for i in range(2):
model = random_model(2)
obs_distns = [
d.Categorical(K=3, alpha_0=1.),
d.Categorical(K=3, alpha_0=1.)
]
stateseq = np.random.randint(2, size=10)
data = np.array([obs_distns[a].rvs() for a in stateseq])
target_val = compute_likelihood_enumeration(obs_distns=obs_distns,
data=data,
**model)
yield make_nose_tuple(_likelihood_helper,
target_val=target_val,
data=data,
obs_distns=obs_distns,
**model)
# Gaussian data
for i in range(2):
model = random_model(3)
obs_distns = [
d.Gaussian(mu=np.random.randn(2), sigma=np.eye(2)),
d.Gaussian(mu=np.random.randn(2), sigma=np.eye(2)),
d.Gaussian(mu=np.random.randn(2), sigma=np.eye(2))
]
stateseq = np.random.randint(3, size=10)
data = np.vstack([obs_distns[a].rvs() for a in stateseq])
target_val = compute_likelihood_enumeration(obs_distns=obs_distns,
data=data,
**model)
yield make_nose_tuple(_likelihood_helper,
target_val=target_val,
data=data,
obs_distns=obs_distns,
**model)
import numpy as np
from matplotlib import pyplot as plt
from pyhsmm import models, distributions
np.random.seed(0)
np.seterr(invalid='raise')
obs_hypparams = dict(mu_0=np.zeros(2), sigma_0=np.eye(2), kappa_0=0.05, nu_0=5)
### generate data
num_modes = 3
true_obs_distns = [
distributions.Gaussian(**obs_hypparams) for i in range(num_modes)
]
data = np.concatenate(
[true_obs_distns[i % num_modes].rvs(25) for i in range(25)])
## inference!
hmm = models.HMM(obs_distns=[
distributions.Gaussian(**obs_hypparams) for i in range(num_modes * 3)
],
alpha=3.,
init_state_concentration=1.)
hmm.add_data(data)
hmm.meanfield_coordinate_descent_step()
scores = [hmm.meanfield_coordinate_descent_step() for i in range(50)]
scores = np.array(scores)
allseqs = np.array_split(alldata, 250)
datas, heldout = hold_out(allseqs, 0.05)
training_size = sum(data.shape[0] for data in datas)
print('...done!')
print('%d total frames' % sum(data.shape[0] for data in alldata))
print('split into %d training and %d test sequences' %
(len(datas), len(heldout)))
### inference!
Nmax = 20
obs_hypparams = dict(mu_0=np.zeros(2), sigma_0=np.eye(2), kappa_0=0.2, nu_0=5)
hmm = models.HMM(
obs_distns=[distributions.Gaussian(**obs_hypparams) for i in range(Nmax)],
alpha=10.,
init_state_concentration=1.)
scores = []
sgdseq = sgd_passes(tau=0, kappa=0.7, datalist=datas)
for t, (data, rho_t) in progprint(enumerate(sgdseq)):
hmm.meanfield_sgdstep(data, data.shape[0] / training_size, rho_t)
if t % 10 == 0:
scores.append(hmm.log_likelihood(heldout))
plt.figure()
plt.plot(scores)
plt.show()