def _train_single(data, n_components, n_pc, covar_types, verbose, n_jobs, n_iter_search):
model = PGMM(n_components=n_components,
n_pc=n_pc,
covariance_type=covar_types,
verbose=verbose)
single_start = time()
model.fit(data)
single_end = time()
logging.info("Single PGMM took %.2f seconds."
% (single_end - single_start))
return model
def _train_single(data, n_components, n_pc, covar_types, verbose, n_jobs,
n_iter_search):
model = PGMM(n_components=n_components,
n_pc=n_pc,
covariance_type=covar_types,
verbose=verbose)
single_start = time()
model.fit(data)
single_end = time()
logging.info("Single PGMM took %.2f seconds." %
(single_end - single_start))
return model
def train_with_parameters(params, data_broadcast):
data = data_broadcast.value
n_components = params[0]
n_pc = params[1]
covariance_type = covar_type_int[params[2]]
model = PGMM(n_components=n_components,
n_pc=n_pc,
covariance_type=covariance_type, tol=1e-6)
try:
model.fit(data)
return model.bic(data), model
except:
# If failed, return the BIC as infite so the model will be discarted.
return float("inf"), None
def train_with_parameters(params, data_broadcast):
data = data_broadcast.value
n_components = params[0]
n_pc = params[1]
covariance_type = covar_type_int[params[2]]
model = PGMM(n_components=n_components,
n_pc=n_pc,
covariance_type=covariance_type,
tol=1e-6)
try:
model.fit(data)
return model.bic(data), model
except:
# If failed, return the BIC as infite so the model will be discarted.
return float("inf"), None
def _train_random(data, n_components, n_pc, covar_types, verbose, n_jobs,
n_iter_search):
param_distribution = {
'n_components': n_components,
'n_pc': n_pc,
'covariance_type': covar_types,
'verbose': [verbose]
}
random_search = RandomizedSearchCV(PGMM(),
param_distributions=param_distribution,
n_iter=n_iter_search,
scoring=PGMM.bic,
verbose=verbose,
n_jobs=n_jobs)
random_start = time()
random_search.fit(data)
random_end = time()
logging.info(
"RandomSearchCV took %.2f seconds for %d candidate parameter settings."
% (random_end - random_start, len(random_search.grid_scores_)))
return random_search.best_estimator_
def _train_grid(data, n_components, n_pc, covar_types, verbose, n_jobs,
n_iter_search):
param_grid = [{
'n_components': n_components,
'n_pc': n_pc,
'covariance_type': covar_types,
'verbose': [verbose]
}]
grid_search = GridSearchCV(PGMM(),
param_grid=param_grid,
scoring=PGMM.bic,
verbose=verbose,
n_jobs=n_jobs)
grid_start = time()
grid_search.fit(data)
grid_end = time()
logging.info(
'GridSearchCV took %.2f seconds for %d candidate parameter settings.' %
(grid_end - grid_start, len(grid_search.grid_scores_)))
return grid_search.best_estimator_