def train_model(data: np.ndarray,
clusters: int = 5,
init_nodes: list = None) -> BayesianNetwork:
bn = BayesNet()
#Сluster the initial data in order to fill in a hidden variable based on the distribution of clusters
kmeans = KMeans(n_clusters=clusters, random_state=0).fit(data)
labels = kmeans.labels_
hidden_dist = DiscreteDistribution.from_samples(labels)
hidden_var = np.array(hidden_dist.sample(data.shape[0]))
new_data = np.column_stack((data, hidden_var))
latent = (new_data.shape[1]) - 1
#Train the network structure on data taking into account a hidden variable
bn = hc_rr(new_data, latent=latent, init_nodes=init_nodes)
structure = []
nodes = sorted(list(bn.nodes()))
for rv in nodes:
structure.append(tuple(bn.F[rv]['parents']))
structure = tuple(structure)
bn = BayesianNetwork.from_structure(new_data, structure)
bn.bake()
#Learn a hidden variable
hidden_var = np.array([np.nan] * (data.shape[0]))
new_data = np.column_stack((data, hidden_var))
bn.predict(new_data)
bn.fit(new_data)
bn.bake()
return (bn)
def mmhc(data, alpha=0.05, metric='AIC', max_iter=100, method='hc'):
"""
Max-Min Hill Climbing Algorithm for
learning a Bayesian Network structure
from data.
Arguments
---------
*data* : a numpy ndarray
*alpha* : a float
Probability of Type II Error for
independence tests.
*metric* : a string
*metric* : a string
Which score metric to use.
Options:
- 'AIC'
- 'BIC'
- 'LL' (log-likelihood)
*method* : a string
The type of hill-climbing algorithm to run
OPTIONS:
- 'hc' : normal hill-climbing
- 'rr' : hill-climbing with random restarts
- 'tabu' : tabu hill-climbing
Returns
-------
*bn* : a BayesNet object
"""
# GET EDGE RESTRICTIONS FROM MMPC
PC_dict = mmpc(data)
restriction = []
for y, pc in PC_dict.items():
for x in pc:
restriction.append((y, x))
# RUN HILL-CLIMBING WITH EDGE RESTRICTIONS
if method == 'tabu':
bn = tabu(data=data,
metric=metric,
max_iter=max_iter,
restriction=restriction)
elif method == 'rr':
bn = hc_rr(data=data,
metric=metric,
max_iter=max_iter,
restriction=restriction)
else:
bn = hc(data=data,
metric=metric,
max_iter=max_iter,
restriction=restriction)
return bn
def mdbn(data, f_cols, c_cols, f_struct='DAG', c_struct='DAG', wrapper=False): """ Learn the structure of a Multi-Dimensional Bayesian Network - typically used for Classification. Note that this structure does not have to be used for classification, since it simply returns a Bayesian Network - albeit with a more unqiue structure than tradiitonally found. If there are any other applications of this bipartite-like BN structure learning, this algorithm can certainly be used. """ f_data = data[:,f_cols] c_data = data[:,c_cols] f_bn = hc_rr(f_data) c_bn = hc_rr(c_data) mbc_bn = bridge(c_bn=c_bn, f_bn=f_bn, data=data) return mbc_bn
def bridge(c_bn, f_bn, data): """ Make a Multi-Dimensional Bayesian Network by bridging two Bayesian network structures. This happens by placing edges from c_bn -> f_bn using a heuristic optimization procedure. This can be used to create a Multi-Dimensional Bayesian Network classifier from two already-learned Bayesian networks - one of which is a BN containing all the class variables, the other containing all the feature variables. Arguments --------- *c_bn* : a BayesNet object with known structure *f_bn* : a BayesNet object with known structure. Returns ------- *m_bn* : a merged/bridge BayesNet object, whose structure contains *c_bn*, *f_bn*, and some bridge edges between them. """ restrict = [] for u in c_bn: for v in f_bn: restrict.append((u,v)) # only allow edges from c_bn -> f_bn bridge_bn = hc_rr(data, restriction=restrict) m_bn = bridge_bn.E m_bn.update(c_bn.E) m_bn.update(f_bn.E) mbc_bn = BayesNet(E=m_bn)
