def test_structure_error():
"""Test structure_error function."""
a = np.eye(3) + np.eye(3, k=1)
b = np.eye(3, k=-1) + np.eye(3)
result = {
'accuracy': 0.5555555555555556,
'average_precision': 0.66666666666666663,
'balanced_accuracy': 0.55,
'dor': 1.4999999999999998,
'f1': 0.6,
'fall_out': 0.5,
'false_omission_rate': 0.5,
'fdr': 0.4,
'fn': 2,
'fp': 2,
'mcc': 0.0,
'miss_rate': 0.4,
'nlr': 0.8,
'npv': 0.5,
'plr': 1.2,
'precision': 0.6,
'prevalence': 0.5555555555555556,
'recall': 0.6,
'specificity': 0.5,
'tn': 2,
'tp': 3
}
assert_equal(utils.structure_error(a, b), result)
b = np.eye(3) + np.eye(3, k=-1) * 1e-3
result = {
'accuracy': 0.7777777777777778,
'average_precision': 0.66666666666666663,
'balanced_accuracy': 0.8,
'dor': 0.0,
'f1': 0.7499999999999999,
'fall_out': 0.0,
'false_omission_rate': 0.3333333333333333,
'fdr': 0.0,
'fn': 2,
'fp': 0,
'miss_rate': 0.4,
'mcc': 0.0,
'nlr': 0.4,
'npv': 0.6666666666666666,
'plr': 0,
'precision': 1.0,
'prevalence': 0.5555555555555556,
'recall': 0.6,
'specificity': 1.0,
'tn': 4,
'tp': 3
}
assert_equal(utils.structure_error(a, b, thresholding=True, eps=1e-2),
result)
def test_structure_error():
"""Test structure_error function."""
a = np.eye(3) + np.eye(3, k=1)
b = np.eye(3, k=-1) + np.eye(3)
result = {
"accuracy": 0.5555555555555556,
"average_precision": 0.66666666666666663,
"balanced_accuracy": 0.55,
"dor": 1.4999999999999998,
"f1": 0.6,
"fall_out": 0.5,
"false_omission_rate": 0.5,
"fdr": 0.4,
"fn": 2,
"fp": 2,
"mcc": 0.0,
"miss_rate": 0.4,
"nlr": 0.8,
"npv": 0.5,
"plr": 1.2,
"precision": 0.6,
"prevalence": 0.5555555555555556,
"recall": 0.6,
"specificity": 0.5,
"tn": 2,
"tp": 3,
}
assert_equal(utils.structure_error(a, b), result)
b = np.eye(3) + np.eye(3, k=-1) * 1e-3
result = {
"accuracy": 0.7777777777777778,
"average_precision": 0.66666666666666663,
"balanced_accuracy": 0.8,
"dor": 0.0,
"f1": 0.7499999999999999,
"fall_out": 0.0,
"false_omission_rate": 0.3333333333333333,
"fdr": 0.0,
"fn": 2,
"fp": 0,
"miss_rate": 0.4,
"mcc": 0.0,
"nlr": 0.4,
"npv": 0.6666666666666666,
"plr": 0,
"precision": 1.0,
"prevalence": 0.5555555555555556,
"recall": 0.6,
"specificity": 1.0,
"tn": 4,
"tp": 3,
}
assert_equal(utils.structure_error(a, b, thresholding=True, eps=1e-2), result)
def tgl_results(X, y, K, K_obs, ells, **params):
mdl = TimeGraphicalLasso(assume_centered=0,
verbose=0,
rtol=1e-5,
tol=1e-5,
max_iter=500,
rho=1. / np.sqrt(X.shape[0]))
tic = time.time()
ll = mdl.set_params(**params).fit(X, y)
tac = time.time()
iterations = ll.n_iter_
F1score = utils.structure_error(K, ll.precision_)['f1']
MSE_observed = None # utils.error_norm(K_obs, ll.precision_ - ll.latent_)
MSE_precision = utils.error_norm(K, ll.precision_)
MSE_latent = None # utils.error_norm(ells, ll.latent_)
mean_rank_error = None # utils.error_rank(ells, ll.latent_)
res = dict(n_dim_obs=K.shape[1],
time=tac - tic,
iterations=iterations,
F1score=F1score,
MSE_precision=MSE_precision,
MSE_observed=MSE_observed,
MSE_latent=MSE_latent,
mean_rank_error=mean_rank_error,
likelihood=mdl.score(X, y),
note=None,
estimator=ll)
return res
def chandresekeran_results(data_grid, K, K_obs, ells, tau, alpha, **kwargs):
emp_cov = np.array([
empirical_covariance(x, assume_centered=True)
for x in data_grid.transpose(2, 0, 1)
]).transpose(1, 2, 0)
rho = 1. / np.sqrt(data_grid.shape[0])
result = lvglasso(emp_cov, alpha, tau, rho)
ma_output = Bunch(**result)
R = np.array(ma_output.R).T
S = np.array(ma_output.S).T
L = np.array(ma_output.L).T
ss = utils.structure_error(K, S)
MSE_observed = utils.error_norm(K_obs, R)
MSE_precision = utils.error_norm(K, S, upper_triangular=True)
MSE_latent = utils.error_norm(ells, L)
mean_rank_error = utils.error_rank(ells, L)
res = dict(n_dim_obs=K.shape[1],
time=ma_output.elapsed_time,
iterations=np.max(ma_output.iter),
MSE_precision=MSE_precision,
MSE_observed=MSE_observed,
MSE_latent=MSE_latent,
mean_rank_error=mean_rank_error,
note=None,
estimator=ma_output,
likelihood=likelihood_score(data_grid.transpose(2, 0, 1), R),
latent=L)
res = dict(res, **ss)
return res
def friedman_results(data_grid, K, K_obs, ells, alpha):
from rpy2.robjects.packages import importr
glasso = importr('glasso').glasso
tic = time.time()
iters = []
precisions = []
for d in data_grid.transpose(2, 0, 1):
emp_cov = empirical_covariance(d)
out = glasso(emp_cov, alpha)
iters.append(int(out[-1][0]))
precisions.append(np.array(out[1]))
tac = time.time()
iterations = np.max(iters)
precisions = np.array(precisions)
F1score = utils.structure_error(K, precisions)['f1']
MSE_observed = None
MSE_precision = utils.error_norm(K, precisions, upper_triangular=True)
MSE_latent = None
mean_rank_error = None
res = dict(n_dim_obs=K.shape[1],
time=tac - tic,
iterations=iterations,
F1score=F1score,
MSE_precision=MSE_precision,
MSE_observed=MSE_observed,
MSE_latent=MSE_latent,
mean_rank_error=mean_rank_error,
likelihood=likelihood_score(data_grid.transpose(2, 0, 1),
precisions),
note=None,
estimator=None)
return res
def glasso_results(data_grid, K, K_obs, ells, alpha):
gl = GLsk(alpha=alpha, mode='cd', assume_centered=False, max_iter=500)
tic = time.time()
iters = []
precisions = []
for d in data_grid.transpose(2, 0, 1):
gl.fit(d)
iters.append(gl.n_iter_)
precisions.append(gl.precision_)
tac = time.time()
iterations = np.max(iters)
precisions = np.array(precisions)
ss = utils.structure_error(K, precisions) #, thresholding=1, eps=1e-5)
MSE_observed = None
MSE_precision = utils.error_norm(K, precisions, upper_triangular=True)
MSE_latent = None
mean_rank_error = None
res = dict(n_dim_obs=K.shape[1],
time=tac - tic,
iterations=iterations,
MSE_precision=MSE_precision,
MSE_observed=MSE_observed,
MSE_latent=MSE_latent,
mean_rank_error=mean_rank_error,
likelihood=likelihood_score(data_grid.transpose(2, 0, 1),
precisions),
note=None,
estimator=gl)
res = dict(res, **ss)
return res
def ltgl_results(X, y, K, K_obs, ells, **params):
mdl = LatentTimeGraphicalLasso(assume_centered=0,
verbose=0,
rtol=1e-5,
tol=1e-5,
max_iter=1000,
rho=1. / np.sqrt(X.shape[0]),
update_rho_options=dict(mu=5))
tic = time.time()
ll = mdl.set_params(**params).fit(X, y)
tac = time.time()
iterations = ll.n_iter_
ss = utils.structure_error(K, ll.precision_) #, thresholding=1, eps=1e-5)
MSE_observed = utils.error_norm(K_obs, ll.precision_ - ll.latent_)
MSE_precision = utils.error_norm(K, ll.precision_, upper_triangular=True)
MSE_latent = utils.error_norm(ells, ll.latent_)
mean_rank_error = utils.error_rank(ells, ll.latent_)
res = dict(n_dim_obs=K.shape[1],
time=tac - tic,
iterations=iterations,
MSE_precision=MSE_precision,
MSE_observed=MSE_observed,
MSE_latent=MSE_latent,
mean_rank_error=mean_rank_error,
note=None,
estimator=ll,
likelihood=mdl.score(X, y),
latent=ll.latent_)
res = dict(res, **ss)
return res
def base_results(mdl, X, y, K, K_obs, ells, search_spaces=None, **params):
ll = mdl.set_params(**params)
tic = time.time()
if search_spaces is None:
ll.fit(X, y)
else:
ll = use_bscv(ll, search_spaces, X, y)
tac = time.time()
ss = utils.structure_error(K, ll.precision_)
MSE_precision = utils.error_norm(K, ll.precision_, upper_triangular=True)
res = dict(n_dim_obs=K.shape[1],
time=tac - tic,
iterations=ll.n_iter_,
MSE_precision=MSE_precision,
estimator=ll,
likelihood=ll.score(X, y))
if hasattr(ll, 'latent_'):
res['MSE_observed'] = utils.error_norm(K_obs,
ll.precision_ - ll.latent_)
res['MSE_latent'] = utils.error_norm(ells, ll.latent_)
res['mean_rank_error'] = utils.error_rank(ells, ll.latent_)
res = dict(res, **ss)
return res
def wp_results(data_list, K, **params):
n_iter = 1000
mdl = wishart_process_.WishartProcess(verbose=True,
n_iter=n_iter,
**params)
X = np.vstack(data_list)
y = np.array([np.ones(x.shape[0]) * i
for i, x in enumerate(data_list)]).flatten().astype(int)
tic = time.time()
ll = mdl.fit(X, y)
tac = time.time()
# mdl.likelihood(wp.D_map)
# mdl.loglikes_after_burnin.max()
mdl.store_precision = True
ss = utils.structure_error(K, ll.precision_, thresholding=False, eps=1e-3)
MSE_precision = utils.error_norm(K, ll.precision_, upper_triangular=True)
res = dict(n_dim_obs=K.shape[1],
time=tac - tic,
iterations=n_iter,
MSE_precision=MSE_precision,
estimator=ll,
likelihood=ll.score(X, y))
res = dict(res, **ss)
return res
def hallac_results(data_grid,
K,
K_obs,
ells,
beta,
alpha,
penalty=2,
tvgl_path=''):
if tvgl_path:
import sys
sys.path.append(tvgl_path)
import TVGL
# with suppress_stdout():
tic = time.time()
thetaSet, empCovSet, status, gvx = TVGL.TVGL(np.vstack(
data_grid.transpose(2, 0, 1)),
data_grid.shape[0],
lamb=alpha,
beta=beta,
indexOfPenalty=penalty)
tac = time.time()
if status != "Optimal":
print("not converged")
precisions = np.array(thetaSet)
ss = utils.structure_error(K, precisions)
MSE_observed = None
MSE_precision = utils.error_norm(K, precisions, upper_triangular=True)
MSE_latent = None
mean_rank_error = None
res = dict(n_dim_obs=K.shape[1],
time=tac - tic,
iterations=gvx.n_iter_,
MSE_precision=MSE_precision,
MSE_observed=MSE_observed,
MSE_latent=MSE_latent,
mean_rank_error=mean_rank_error,
likelihood=likelihood_score(data_grid.transpose(2, 0, 1),
precisions),
note=status,
estimator=gvx)
res = dict(res, **ss)
return res
def set_results(
vs, model, name, i, labels_true, labels_pred, thetas_true_sparse,
thetas_true_rep, obs_precs_sparse, obs_precs, tac):
th = name in ['wp', 'ticc']
vs.setdefault((name, i), {}).setdefault('model', []).append(model)
vs.setdefault(
(name, i),
{}).setdefault('v_measure',
[]).append(v_measure_score(labels_true, labels_pred))
vs.setdefault((name, i), {}).setdefault('structure_error', []).append(
structure_error(
thetas_true_sparse, obs_precs_sparse, no_diagonal=True,
thresholding=th, eps=1e-5))
vs.setdefault(
(name, i),
{}).setdefault('error_norm',
[]).append(error_norm_time(thetas_true_rep, obs_precs))
vs.setdefault((name, i), {}).setdefault('error_norm_sparse', []).append(
error_norm_time(thetas_true_sparse, obs_precs_sparse))
vs.setdefault((name, i), {}).setdefault('time', []).append(tac)
def lgl_results(data_grid, K, K_obs, ells, **params):
mdl = LatentGraphicalLasso(assume_centered=0,
verbose=0,
rtol=1e-5,
tol=1e-5,
max_iter=500,
rho=1. / np.sqrt(data_grid.shape[0]))
tic = time.time()
iters = []
precisions, latents = [], []
for d in data_grid.transpose(2, 0, 1):
mdl.set_params(**params).fit(d)
iters.append(mdl.n_iter_)
precisions.append(mdl.precision_)
latents.append(mdl.latent_)
tac = time.time()
iterations = np.max(iters)
precisions = np.array(precisions)
latents = np.array(latents)
F1score = utils.structure_error(K, precisions)['f1']
MSE_observed = utils.error_norm(K_obs, precisions - latents)
MSE_precision = utils.error_norm(K, precisions)
MSE_latent = utils.error_norm(ells, latents)
mean_rank_error = utils.error_rank(ells, latents)
res = dict(n_dim_obs=K.shape[1],
time=tac - tic,
iterations=iterations,
F1score=F1score,
MSE_precision=MSE_precision,
MSE_observed=MSE_observed,
MSE_latent=MSE_latent,
mean_rank_error=mean_rank_error,
note=None,
estimator=mdl)
return res