def testTranslationAbility():
#
min_lengths = np.array((20,))
num_start_times = 5
num_lengths = 1
num_examples = 100
min_prob = 0.01
(E,
min_lengths,
example_labeled_start_times,
background_model) = getTestingDefaults(min_lengths,
num_start_times,
num_lengths,
num_examples)
# run through the experiment step by step
num_examples = len(E)
num_models = len(min_lengths)
max_loglikes_by_example = np.zeros(num_examples)
log_background_model = np.log(background_model)
log_inv_background_model = np.log(1-background_model)
L,example_lls,M,M_normalizations,weights = EM_loc_length.initSuffStats(E,
num_models,
num_start_times,
num_lengths,
min_prob=min_prob)
M_normalizations[:] = 0
M, weights = EM_loc_length.MStep(E,L,min_lengths,
example_labeled_start_times,
M,M_normalizations,weights,
example_lls,min_prob=min_prob
)
(E_bgd_frame_scores,
E_bgd_scores_prefix,
E_all_bgd_scores) = EM_loc_length.computeBackgroundLogLikelihoods(E,log_background_model,log_inv_background_model)
log_like,L = EM_loc_length.EStep(E,L,min_lengths,
example_labeled_start_times,np.log(weights),np.log(M),
np.log(1-M),max_loglikes_by_example,E_bgd_frame_scores,
E_bgd_scores_prefix,E_all_bgd_scores,example_lls)
M_normalizations[:] = 0
M, weights = EM_loc_length.MStep(E,L,min_lengths,
example_labeled_start_times,
M,M_normalizations,weights,
example_lls,min_prob=min_prob
)
M,weights,L = EM_loc_length.EM(E,.0001,min_lengths,example_labeled_start_times,
num_start_times,num_lengths,background_model,
min_prob=0.01,verbose=True)
def testSimpleComputeLogLikelihood():
np.random.seed(0)
min_lengths = np.array((20,))
num_features=100
num_lengths=1
num_ridges=4
min_probability=0.01
M = createRandomModels(min_lengths=min_lengths,
num_features=num_features,
num_lengths=num_lengths,
num_ridges=num_ridges,
blurring_bandwidth=2,
min_probability=min_probability
)
log_M = np.log(M)
log_inv_M = np.log(1-M)
num_models = len(M)
background_model = M.min() * np.ones(num_features)
# generate the data set
# num_start_times specifies how many start times are allowed
# for the models
num_start_times = 1
num_examples=10
E,example_labeled_start_times = createRandomData(models=M,
num_examples=num_examples,
min_lengths=min_lengths,
num_lengths=num_lengths,
num_start_times=num_start_times,
background_model=background_model,
)
L = np.zeros((num_examples,num_models,num_start_times,
num_lengths))
max_loglikes_by_example = np.zeros(num_examples)
(E_bgd_frame_scores,
E_bgd_scores_prefix,
E_all_bgd_scores) = EM_loc_length.computeBackgroundLogLikelihoods(E,np.log(background_model),np.log(1-background_model))
log_weights = np.log(np.ones((num_models,num_lengths)) * 1./(num_models * num_lengths))
EM_loc_length.computeLogLikelihood(E,L,
min_lengths,
example_labeled_start_times,
log_weights,
log_M,
log_inv_M,
max_loglikes_by_example,
E_bgd_frame_scores,
E_bgd_scores_prefix,E_all_bgd_scores)
example_lls = np.zeros((num_examples,num_lengths))
L[:] = 0.
out_ll = EM_loc_length.EStep(E,L,min_lengths,
example_labeled_start_times,log_weights,log_M,
log_inv_M,max_loglikes_by_example, E_bgd_frame_scores,
E_bgd_scores_prefix,E_all_bgd_scores,example_lls)
weights = np.zeros(log_weights.shape)
M_normalizations = np.zeros(log_M.shape)
M[:] = 0
EM_loc_length.MStep(E,L,min_lengths,
example_labeled_start_times,M,M_normalizations,weights,
example_lls
)
