def train_HMM(n_components,
observations,
lengths,
labels,
verbose=False,
cov_type='spherical',
iter=500):
AD_idx = np.where(labels == 1)
AD_lengths = lengths[AD_idx[0]]
nonAD_idx = np.where(labels != 1)
nonAD_lengths = lengths[nonAD_idx[0]]
l = np.nditer(labels)
nonAD_observations = np.zeros((0, observations.shape[1]))
AD_observations = np.zeros((0, observations.shape[1]))
for s, e, in iter_from_X_lengths(observations, lengths):
temp = observations[s:e, :]
if l.next() == 0:
nonAD_observations = np.vstack((nonAD_observations, temp))
else:
AD_observations = np.vstack((AD_observations, temp))
AD_hmm = cf.setup_and_train(n_components, AD_observations, AD_lengths,
cov_type, verbose, iter)
nonAD_hmm = cf.setup_and_train(n_components, nonAD_observations,
nonAD_lengths, cov_type, verbose, iter)
return AD_hmm, nonAD_hmm
def test_HMM(ad_hmm, non_ad_hmm, observations, lengths):
""" Test observation probabilities for both HMMs and decide on label """
predicted = []
for start, end in iter_from_X_lengths(observations, lengths):
obs = observations[start:end, :]
ad_score = ad_hmm.score(obs)
non_ad_score = non_ad_hmm.score(obs)
if ad_score > non_ad_score:
predicted.append(1)
else:
predicted.append(0)
return predicted
from hmmlearn import hmm
import numpy as np
from common_functions import iter_from_X_lengths
from common_functions import random_ints_with_sum
np.random.seed(42)
model = hmm.GaussianHMM(n_components=3, covariance_type="full")
model.startprob_ = np.array([0.6, 0.3, 0.1])
model.transmat_ = np.array([[0.7, 0.2, 0.1],
[0.3, 0.5, 0.2],
[0.3, 0.3, 0.4]])
model.means_ = np.array([[0.0, 0.0], [3.0, -3.0], [5.0, 10.0]])
model.covars_ = np.tile(np.identity(2), (3, 1, 1))
for i in np.arange(1):
obs, _ = model.sample(10000)
model.fit(obs)
N = 100
obs, _ = model.sample(N)
lengths = list(random_ints_with_sum(N))
for i, j in iter_from_X_lengths(obs, lengths):
observation = obs[i:j, :]
logprob = model.score(observation)
logprob1 = model.score(observation, [observation.shape[0]])
print(logprob, logprob1)
print(observation.shape[0], j - i)
fpr_3 = {}
tpr_3 = {}
for f in SVM_folds:
sensitivity_3[f] = [0] * (n_components[-1] + 1)
specificity_3[f] = [0] * (n_components[-1] + 1)
CM_3[f] = {}
for n_comp in n_components:
CM_3[f][n_comp] = np.zeros((2, 2))
""" Make train/test data """
train_obs_init, train_len_init, train_labels_init, MCI_obs, MCI_len, MCI_labels = \
train_HMM.make_data()
starts = []
ends = []
for s, e in iter_from_X_lengths(MCI_obs, MCI_len):
starts.append(s)
ends.append(e)
if folds != 1:
skf = StratifiedKFold(MCI_labels, 3)
else:
skf = zip([-1], [-1])
for iteration in np.arange(iterations):
f = 0
for train, test in skf:
if np.any(train == -1):
train_len = train_len_init
train_labels = train_labels_init
train_obs = train_obs_init.copy()