def _run_alg(excluded):
setting_list_exclude = [setting for setting in setting_list if excluded not in setting['known_interventions']]
iv_estimated_folder = os.path.join(ESTIMATED_FOLDER, 'exclude_%s' % excluded)
filename = os.path.join(iv_estimated_folder,
'utigsp_%s_alpha=%.2e,alpha_i=%.2e.txt' % (ci_test, alpha, alpha_invariance))
if not os.path.exists(filename):
if ci_test == 'gauss_ci':
os.system('touch gauss_ci_utigsp_%d.tst' % excluded)
est_dag = unknown_target_igsp(
obs_samples,
setting_list_exclude,
suffstat,
nnodes,
gauss_ci_test,
hsic_invariance_test,
alpha=alpha,
nruns=10,
alpha_invariance=alpha_invariance,
)
elif ci_test == 'hsic':
os.system('touch hsic_utigsp_%d.tst' % excluded)
est_dag = unknown_target_igsp(
obs_samples,
setting_list_exclude,
obs_samples,
nnodes,
hsic_test,
hsic_invariance_test,
alpha=alpha,
nruns=10,
alpha_invariance=alpha_invariance,
)
else:
raise ValueError
np.savetxt(filename, est_dag.to_amat())
np.savetxt(os.path.join(sample_folder, 'observational.txt'), samples)
iv_str = 'known_ivs=%s;unknown_ivs=%s.txt' % (nnodes - 1, 0)
np.savetxt(os.path.join(iv_sample_folder, iv_str), iv_samples)
# === RUN ALGORITHMS
dag_igsp = igsp(sample_dict,
suffstat,
nnodes,
gauss_ci_test,
hsic_invariance_test,
nruns=10,
alpha_invariance=1e-5)
dag_utigsp = unknown_target_igsp(sample_dict,
suffstat,
nnodes,
gauss_ci_test,
hsic_invariance_test,
nruns=10,
alpha_invariance=1e-2)
amat_gies = run_gies(sample_folder, lambda_=50)
dag_gies = cd.DAG.from_amat(amat_gies)
dags_igsp.append(dag_igsp)
dags_utigsp.append(dag_utigsp)
dags_gies.append(dag_gies)
shds_igsp.append(dag_igsp.shd(d))
shds_utigsp.append(dag_utigsp.shd(d))
shds_gies.append(dag_gies.shd(d))
print(np.mean(shds_igsp))
print(np.mean(shds_utigsp))
# === RUN IGSP
igsp_dag = igsp(sample_dict,
suffstat,
nnodes,
gauss_ci_test,
hsic_invariance_test,
alpha=1e-5,
alpha_invariance=1e-5,
depth=6,
verbose=True,
starting_permutations=starting_permutations)
np.random.seed(1729)
random.seed(1729)
# === RUN UTIGSP
utigsp_dag = unknown_target_igsp(obs_samples,
setting_list,
suffstat,
nnodes,
gauss_ci_test,
hsic_invariance_test,
alpha=1e-5,
alpha_invariance=1e-5,
depth=6,
verbose=True,
starting_permutations=starting_permutations)
print(true_dag.shd(utigsp_dag))
print(true_dag.shd(igsp_dag))
print(igsp_dag.shd(utigsp_dag))
import random
np.random.seed(40)
random.seed(9879132)
nnodes = 10
nsamples = 100
dag = cd.rand.directed_erdos(nnodes, 1.5 / (nnodes - 1), 1)
gdag = cd.rand.rand_weights(dag)
obs_samples = gdag.sample(nsamples)
setting_list = []
for i in range(10):
iv_samples = gdag.sample_interventional_perfect(
{
i: cd.GaussIntervention(1, .1),
0: cd.GaussIntervention(1, .1)
}, nsamples)
setting_list.append({'known_interventions': {i}, 'samples': iv_samples})
suffstat = dict(C=np.corrcoef(obs_samples, rowvar=False), n=nsamples)
est_dag, learned_intervention_targets = unknown_target_igsp(
obs_samples,
setting_list,
suffstat,
nnodes,
gauss_ci_test,
hsic_invariance_test,
1e-5,
1e-5,
nruns=5,
verbose=True)