def test_pairwise_causal_estimation():
np.random.seed(0)
n = 6
data, M = create_data(n, 20000, enb=2)
dags, scores = sumu.Gadget(data=data, mcmc={"iters": 80000}).sample()
causal_effects = sumu.beeps(dags, data)
mse = ((np.linalg.inv(np.eye(n) - M["B"]) - causal_effects.mean(axis=0))**2).mean()
print(mse)
assert mse < 0.1
def test_Gadget_runs_n_between_193_and_256():
# NOTE: This does not test all numbers of variables between 193 and 256
data_path = pathlib.Path(__file__).resolve().parents[2] / "data"
bn_path = data_path / "andes.dsc"
bn = sumu.utils.io.read_dsc(bn_path)
data = bn.sample(200, seed=0)
g = sumu.Gadget(data=data, cp_algo="top", K=10, d=2, mc3=2, burn_in=100, iterations=100, thinning=2)
sumu.utils.io.pretty_dict(g.params)
g.sample()
assert True
def test_Gadget_runs_n_between_129_and_192():
# NOTE: This does not test all numbers of variables between 129 and 192
data_path = pathlib.Path(__file__).resolve().parents[2] / "data"
bn_path = data_path / "munin1.dsc"
bn = sumu.utils.io.read_dsc(bn_path)
data = bn.sample(200, seed=0)
g = sumu.Gadget(data=data,
mcmc={"iters": 200, "mc3": 2, "burn_in": 0.5, "n_dags": 50},
candp={"name": "top"}, cons={"K": 10, "d": 2})
g.sample()
assert True
def test_Gadget_empirical_edge_prob_error_decreases():
params = {
# generic MCMC parameters
"mcmc": {"n_indep": 1, "iters": 150000,
"mc3": 3, "burn_in": 0.5, "n_dags": 10000},
# score to use and its parameters
"score": {"name": "bdeu", "params": {"ess": 10}},
# modular structure prior and its parameters
"prior": {"name": "fair"},
# constraints on the DAG space
"cons": {
"max_id": -1,
"K": 8,
"d": 3,
"pruning_eps": 0.001
},
# algorithm to use for finding candidate parents
"candp": {"name": "greedy-lite", "params": {"k": 6}},
# preparing for catastrofic cancellations
"catc": {
"tolerance": 2**-32,
"cache_size": 10**7
},
# Logging
"logging": {
"stats_period": 15,
#"logfile": "gadget.log"
}
}
data_path = pathlib.Path(__file__).resolve().parents[2] / "data"
bn_path = data_path / "sachs.dsc"
bn = sumu.utils.io.read_dsc(bn_path)
data = bn.sample(200, seed=0)
ls = sumu.gadget.LocalScore(data=data, maxid=-1, score=params["score"])
pset_probs = sumu.aps(ls.candidate_scores(), as_dict=True)
edge_probs = sumu.utils.edge_probs_from_pset_probs(pset_probs)
g = sumu.Gadget(data=data, **params)
dags, scores = g.sample()
max_errors = sumu.utils.utils.edge_empirical_prob_max_error(dags,
edge_probs)
print(max_errors[-1])
assert max_errors[-1] < 0.05
def test_Gadget_empirical_edge_prob_error_decreases():
params = {
# score to use and its parameters
"score": {"name": "bdeu", "ess": 10},
# modular structure prior and its parameters
"prior": {"name": "fair"},
# constraints on the DAG space
"max_id": -1,
"K": 8,
"d": 3,
# algorithm to use for finding candidate parents
"cp_algo": "greedy-lite",
# generic MCMC parameters
"mc3": 2,
"burn_in": 50000,
"iterations": 50000,
"thinning": 5,
# preparing for catastrofic cancellations
"cc_tolerance": 2**-32,
"cc_cache_size": 10**7,
# pruning candidate parent sets
"pruning_eps": 0.001
}
data_path = pathlib.Path(__file__).resolve().parents[2] / "data"
bn_path = data_path / "sachs.dsc"
bn = sumu.utils.io.read_dsc(bn_path)
data = bn.sample(200, seed=0)
ls = sumu.gadget.LocalScore(data=data, maxid=-1, score=params["score"])
pset_probs = sumu.aps(ls.all_candidate_restricted_scores(), as_dict=True)
edge_probs = sumu.utils.edge_probs_from_pset_probs(pset_probs)
# To set the seed for MCMC
g = sumu.Gadget(data=data, **params)
sumu.utils.io.pretty_dict(g.params)
dags, scores = g.sample()
max_errors = sumu.utils.utils.edge_empirical_prob_max_error(dags,
edge_probs)
print(max_errors[-1])
assert max_errors[-1] < 0.05
arr = np.array(df)
data = sumu.Data(arr)
params = {
"array": arr,
"data": data,
"scoref": "bge", # Or "bdeu" for discrete data.
# "ess": 10, # If using BDeu.
"max_id": -1,
"K": 7,
"d": 10,
"cp_algo": "greedy-lite",
"mc3_chains": 10,
"burn_in": 100,
"iterations": 5000,
"thinning": 10
}
g = sumu.Gadget(data=data,
array=arr,
burn_in=100,
iterations=5000,
thinning=10)
dag = g.sample()
end = time.time()
print(dag)
print(end - start)
def test_Gadget_runs_continuous_data():
data = np.random.rand(200, 10)
sumu.Gadget(data=data, cons={"K": 8}).sample()
assert True
params = {
"array": test_arr,
"data": data,
"scoref": "bge", # Or "bdeu" for discrete data.
# "ess": 10, # If using BDeu.
"max_id": -1,
"K": 7,
"d": 5,
"cp_algo": "greedy-lite",
"mc3_chains": 15,
"burn_in": 10,
"iterations": 50000,
"thinning": 10
}
g = sumu.Gadget(**params)
dag = g.sample()
end = time.time()
def dag_validation(dag, result):
relations = dict()
for x in list(dag.nodes):
relations[x] = []
for e in list(dag.edges):
relations[e[0]].append(e[1])
relations_result = dict()
for x in list(dag.nodes):
relations_result[x] = []
for x in relations: