def bayesian_network_datasets(name="asia", samples=10000):
"""
Generates well known sample/toy datasets for bayesian networks,
by sampling from existing graph model.
Parameters
----------
name: str, (default:'asia')
Name of the model to sample from
samples: int, (default: 10000)
Number of observations for our dataset
Returns
-------
pd.DataFrame
"""
model = bn.import_DAG(name)
df = bn.sampling(model, n=samples)
return df
def test_sampling():
# TEST 1:
model = bn.import_DAG('Sprinkler')
n = np.random.randint(10, 1000)
df = bn.sampling(model, n=n)
assert df.shape == (n, 4)
def test_structure_learning():
import bnlearn as bn
df = bn.import_example()
model = bn.structure_learning.fit(df)
assert [*model.keys()] == ['model', 'model_edges', 'adjmat', 'config']
model = bn.structure_learning.fit(df, methodtype='hc', scoretype='bic')
assert [*model.keys()] == ['model', 'model_edges', 'adjmat', 'config']
model = bn.structure_learning.fit(df, methodtype='hc', scoretype='k2')
assert [*model.keys()] == ['model', 'model_edges', 'adjmat', 'config']
model = bn.structure_learning.fit(df, methodtype='cs', scoretype='bdeu')
assert [*model.keys()] == [
'undirected', 'undirected_edges', 'pdag', 'pdag_edges', 'dag',
'dag_edges', 'model', 'model_edges', 'adjmat', 'config'
]
model = bn.structure_learning.fit(df, methodtype='cs', scoretype='k2')
assert [*model.keys()] == [
'undirected', 'undirected_edges', 'pdag', 'pdag_edges', 'dag',
'dag_edges', 'model', 'model_edges', 'adjmat', 'config'
]
model = bn.structure_learning.fit(df, methodtype='ex', scoretype='bdeu')
assert [*model.keys()] == ['model', 'model_edges', 'adjmat', 'config']
model = bn.structure_learning.fit(df, methodtype='ex', scoretype='k2')
assert [*model.keys()] == ['model', 'model_edges', 'adjmat', 'config']
model = bn.structure_learning.fit(df, methodtype='cl', root_node='Cloudy')
assert [*model.keys()] == ['model', 'model_edges', 'adjmat', 'config']
# Test the filtering
DAG = bn.import_DAG('asia')
# Sampling
df = bn.sampling(DAG, n=1000)
# Structure learning of sampled dataset
model = bn.structure_learning.fit(df)
assert np.all(
model['adjmat'].columns.values ==
['smoke', 'bronc', 'lung', 'asia', 'tub', 'either', 'dysp', 'xray'])
# hc Enforce and filtering
model = bn.structure_learning.fit(df,
methodtype='hc',
white_list=['smoke', 'either'],
bw_list_method='filter')
assert np.all(model['adjmat'].columns.values == ['smoke', 'either'])
model = bn.structure_learning.fit(df,
methodtype='hc',
white_list=['smoke', 'either'],
bw_list_method='enforce')
assert np.all(
model['adjmat'].columns.values ==
['smoke', 'bronc', 'lung', 'asia', 'tub', 'either', 'dysp', 'xray'])
model = bn.structure_learning.fit(df,
methodtype='hc',
black_list=['smoke', 'either'],
bw_list_method='filter')
assert np.all(model['adjmat'].columns.values ==
['bronc', 'lung', 'asia', 'tub', 'dysp', 'xray'])
model = bn.structure_learning.fit(df,
methodtype='hc',
scoretype='bic',
black_list=['smoke', 'either'],
bw_list_method='enforce')
assert np.all(
model['adjmat'].columns.values ==
['smoke', 'bronc', 'lung', 'asia', 'tub', 'either', 'dysp', 'xray'])
# hc filter
model = bn.structure_learning.fit(df,
methodtype='ex',
white_list=['smoke', 'either'],
bw_list_method='filter')
assert np.all(model['adjmat'].columns.values == ['either', 'smoke'])
model = bn.structure_learning.fit(
df,
methodtype='ex',
black_list=['asia', 'tub', 'either', 'dysp', 'xray'],
bw_list_method='filter')
assert np.all(model['adjmat'].columns.values == ['bronc', 'lung', 'smoke'])
# cs filter
model = bn.structure_learning.fit(df,
methodtype='cs',
white_list=['smoke', 'either'],
bw_list_method='filter')
assert np.all(model['adjmat'].columns.values == ['either', 'smoke'])
model = bn.structure_learning.fit(
df,
methodtype='cs',
black_list=['asia', 'tub', 'either', 'dysp', 'xray'],
bw_list_method='filter')
assert np.all(model['adjmat'].columns.values == ['bronc', 'smoke', 'lung'])
# cl filter
model = bn.structure_learning.fit(df,
methodtype='cl',
white_list=['smoke', 'either'],
bw_list_method='filter',
root_node='smoke')
assert np.all(model['adjmat'].columns.values == ['smoke', 'either'])
model_ex_bic = bn.structure_learning.fit(df, methodtype='ex', scoretype='bic')
model_ex_k2 = bn.structure_learning.fit(df, methodtype='ex', scoretype='k2')
model_ex_bdeu = bn.structure_learning.fit(df,
methodtype='ex',
scoretype='bdeu')
bn.compare_networks(model, model_hc_bic, pos=G['pos'])
# %% Example with dataset
import bnlearn as bn
DAG = bn.import_DAG('sprinkler')
# Print cpds
bn.print_CPD(DAG)
# plot ground truth
G = bn.plot(DAG)
df = bn.sampling(DAG, n=100)
# %% Inference using custom DAG
import bnlearn as bn
# Load asia DAG
df = bn.import_example('asia')
# from tabulate import tabulate
# print(tabulate(df.head(), tablefmt="grid", headers="keys"))
print(df)
edges = [('smoke', 'lung'), ('smoke', 'bronc'), ('lung', 'xray'),
('bronc', 'xray')]
# edges = [('smoke', 'xray'),
# ('bronc', 'lung')]
# %%
print(dir(bn.structure_learning))
print(dir(bn.parameter_learning))
print(dir(bn.inference))
# %%
# Example dataframe sprinkler_data.csv can be loaded with:
df = bn.import_example()
# df = pd.read_csv('sprinkler_data.csv')
model = bn.structure_learning.fit(df)
G = bn.plot(model)
# %% Load example dataframe from sprinkler
DAG = bn.import_DAG('sprinkler', verbose=0)
df = bn.sampling(DAG, n=1000, verbose=0)
# Structure learning
model = bn.structure_learning.fit(df, verbose=0)
# Plot
G = bn.plot(model)
model_hc_bic = bn.structure_learning.fit(df, methodtype='hc', scoretype='bic', verbose=0)
# %% Load example dataframe from sprinkler
df = bn.import_example('sprinkler', verbose=0)
# Structure learning
model = bn.structure_learning.fit(df, verbose=0)
# Plot
G = bn.plot(model, verbose=0)
# Print the CPDs
bnlearn.print_CPD(DAGnew)
# Make inference
q4 = bnlearn.inference.fit(DAGnew, variables=['bronc','lung'], evidence={'smoke':1, 'xray':0})
# q4 = bnlearn.inference.fit(DAGnew, variables=['bronc','lung'], evidence={'smoke':0, 'xray':0})
# %% Example compare networks
# Load asia DAG
import bnlearn
DAG = bnlearn.import_DAG('asia')
# plot ground truth
G = bnlearn.plot(DAG)
# Sampling
df = bnlearn.sampling(DAG, n=10000)
# Structure learning of sampled dataset
model_sl = bnlearn.structure_learning.fit(df, methodtype='hc', scoretype='bic')
# Plot based on structure learning of sampled data
bnlearn.plot(model_sl, pos=G['pos'])
# Compare networks and make plot
bnlearn.compare_networks(model, model_sl, pos=G['pos'])
# Structure learning with black list
model_wl = bnlearn.structure_learning.fit(df, methodtype='hc', white_list=['asia','tub','bronc','xray','smoke'])
bnlearn.plot(model_wl, pos=G['pos'])
model_bl = bnlearn.structure_learning.fit(df, methodtype='hc', black_list=['asia','tub'])
bnlearn.plot(model_bl, pos=G['pos'])
y_distribution_list.append(y_distribution.get(key, 0))
norm_x = [float(i) / sum(x_distribution_list) for i in x_distribution_list]
norm_y = [float(i) / sum(y_distribution_list) for i in y_distribution_list]
return entropy(norm_x, qk=norm_y)
# Load asia DAG
dataset = "child"
model = bn.import_DAG(dc[dataset]["path"])
# plot ground truth
# G = bn.plot(model)
sample_size = [int(1000 * 100 ** (t / 9)) for t in range(0, 10)]
print(sample_size)
# Sampling
true_df = bn.sampling(model, n=100000)
columns = [col for col in true_df.columns]
print("Get true distribution")
true_distribution = get_distribution(true_df, columns)
kl_divergence = dict()
for size in sample_size:
train_df = true_df.sample(n=size)
# Structure learning of sampled dataset
model_learned = bn.structure_learning.fit(true_df, methodtype='cs', scoretype='bic', verbose=0)
# model_learned = bn.structure_learning.fit(train_df, methodtype='cl', scoretype='bic', root_node='either', verbose=0)
# Parameter learning of sampled dataset
model_learned = bn.parameter_learning.fit(model_learned, train_df, verbose=0)