1
0
Fork 0
Old engine for Continuous Time Bayesian Networks. Superseded by reCTBN. 🐍 https://github.com/madlabunimib/PyCTBN
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
This repo is archived. You can view files and clone it, but cannot push or open issues/pull-requests.
PyCTBN/main_package/classes/structure_estimator.py

253 lines
10 KiB

import itertools
import json
import typing
import networkx as nx
import numpy as np
from networkx.readwrite import json_graph
from scipy.stats import chi2 as chi2_dist
from scipy.stats import f as f_dist
import cache as ch
import conditional_intensity_matrix as condim
import network_graph as ng
import parameters_estimator as pe
import sample_path as sp
import structure as st
class StructureEstimator:
"""
Has the task of estimating the network structure given the trajectories in samplepath.
:_sample_path: the _sample_path object containing the trajectories and the real structure
:_exp_test_sign: the significance level for the exponential Hp test
:_chi_test_alfa: the significance level for the chi Hp test
:_nodes: the _nodes labels
:_nodes_vals: the nodes cardinalities
:_nodes_indxs: the nodes indexes
:_complete_graph: the complete directed graph built using the nodes labels in nodes
:_cache: the _cache object
"""
def __init__(self, sample_path: sp.SamplePath, exp_test_alfa: float, chi_test_alfa: float):
"""
Parameters:
:_sample_path: the _sample_path object containing the trajectories and the real structure
:_exp_test_sign: the significance level for the exponential Hp test
:_chi_test_alfa: the significance level for the chi Hp test
"""
self._sample_path = sample_path
self._nodes = np.array(self._sample_path.structure.nodes_labels)
self._nodes_vals = self._sample_path.structure.nodes_values
self._nodes_indxs = self._sample_path.structure.nodes_indexes
self._complete_graph = self.build_complete_graph(self._sample_path.structure.nodes_labels)
self._exp_test_sign = exp_test_alfa
self._chi_test_alfa = chi_test_alfa
self._cache = ch.Cache()
def build_complete_graph(self, node_ids: typing.List) -> nx.DiGraph:
"""
Builds a complete directed graph (no self loops) given the nodes labels in the list node_ids:
Parameters:
node_ids: the list of nodes labels
Returns:
a complete Digraph Object
"""
complete_graph = nx.DiGraph()
complete_graph.add_nodes_from(node_ids)
complete_graph.add_edges_from(itertools.permutations(node_ids, 2))
return complete_graph
def complete_test(self, test_parent: str, test_child: str, parent_set: typing.List, child_states_numb: int,
tot_vars_count: int) -> bool:
"""
Performs a complete independence test on the directed graphs G1 = test_child U parent_set
G2 = G1 U test_parent (added as an additional parent of the test_child).
Generates all the necessary structures and datas to perform the tests.
Parameters:
test_parent: the node label of the test parent
test_child: the node label of the child
parent_set: the common parent set
child_states_numb: the cardinality of the test_child
tot_vars_count: the total number of variables in the net
Returns:
True iff test_child and test_parent are independent given the sep_set parent_set
False otherwise
"""
p_set = parent_set[:]
complete_info = parent_set[:]
complete_info.append(test_child)
parents = np.array(parent_set)
parents = np.append(parents, test_parent)
sorted_parents = self._nodes[np.isin(self._nodes, parents)]
cims_filter = sorted_parents != test_parent
sofc1 = self._cache.find(set(p_set))
if not sofc1:
bool_mask1 = np.isin(self._nodes, complete_info)
l1 = list(self._nodes[bool_mask1])
indxs1 = self._nodes_indxs[bool_mask1]
vals1 = self._nodes_vals[bool_mask1]
eds1 = list(itertools.product(parent_set,test_child))
s1 = st.Structure(l1, indxs1, vals1, eds1, tot_vars_count)
g1 = ng.NetworkGraph(s1)
g1.fast_init(test_child)
p1 = pe.ParametersEstimator(self._sample_path.trajectories, g1)
p1.fast_init(test_child)
sofc1 = p1.compute_parameters_for_node(test_child)
self._cache.put(set(p_set), sofc1)
sofc2 = None
p_set.insert(0, test_parent)
if p_set:
sofc2 = self._cache.find(set(p_set))
if not sofc2:
complete_info.append(test_parent)
bool_mask2 = np.isin(self._nodes, complete_info)
l2 = list(self._nodes[bool_mask2])
indxs2 = self._nodes_indxs[bool_mask2]
vals2 = self._nodes_vals[bool_mask2]
eds2 = list(itertools.product(p_set, test_child))
s2 = st.Structure(l2, indxs2, vals2, eds2, tot_vars_count)
g2 = ng.NetworkGraph(s2)
g2.fast_init(test_child)
p2 = pe.ParametersEstimator(self._sample_path.trajectories, g2)
p2.fast_init(test_child)
sofc2 = p2.compute_parameters_for_node(test_child)
self._cache.put(set(p_set), sofc2)
for cim1, p_comb in zip(sofc1.actual_cims, sofc1.p_combs):
cond_cims = sofc2.filter_cims_with_mask(cims_filter, p_comb)
for cim2 in cond_cims:
if not self.independence_test(child_states_numb, cim1, cim2):
return False
return True
def independence_test(self, child_states_numb: int, cim1: condim.ConditionalIntensityMatrix,
cim2: condim.ConditionalIntensityMatrix) -> bool:
"""
Compute the actual independence test using two cims.
It is performed first the exponential test and if the null hypothesis is not rejected,
it is permormed also the chi_test.
Parameters:
child_states_numb: the cardinality of the test child
cim1: a cim belonging to the graph without test parent
cim2: a cim belonging to the graph with test parent
Returns:
True iff both tests do NOT reject the null hypothesis of indipendence
False otherwise
"""
M1 = cim1.state_transition_matrix
M2 = cim2.state_transition_matrix
r1s = M1.diagonal()
r2s = M2.diagonal()
C1 = cim1.cim
C2 = cim2.cim
F_stats = C2.diagonal() / C1.diagonal()
exp_alfa = self._exp_test_sign
for val in range(0, child_states_numb):
if F_stats[val] < f_dist.ppf(exp_alfa / 2, r1s[val], r2s[val]) or \
F_stats[val] > f_dist.ppf(1 - exp_alfa / 2, r1s[val], r2s[val]):
#print("CONDITIONALLY DEPENDENT EXP")
return False
M1_no_diag = M1[~np.eye(M1.shape[0], dtype=bool)].reshape(M1.shape[0], -1)
M2_no_diag = M2[~np.eye(M2.shape[0], dtype=bool)].reshape(
M2.shape[0], -1)
chi_2_quantile = chi2_dist.ppf(1 - self._chi_test_alfa, child_states_numb - 1)
Ks = np.sqrt(r1s / r2s)
Ls = np.sqrt(r2s / r1s)
for val in range(0, child_states_numb):
Chi = np.sum(np.power(Ks[val] * M2_no_diag[val] - Ls[val] *M1_no_diag[val], 2) /
(M1_no_diag[val] + M2_no_diag[val]))
if Chi > chi_2_quantile:
return False
return True
def one_iteration_of_CTPC_algorithm(self, var_id: str, tot_vars_count: int):
"""
Performs an iteration of the CTPC algorithm using the node var_id as test_child.
Parameters:
var_id: the node label of the test child
tot_vars_count: the number of _nodes in the net
Returns:
void
"""
print("##################TESTING VAR################", var_id)
u = list(self._complete_graph.predecessors(var_id))
child_states_numb = self._sample_path.structure.get_states_number(var_id)
b = 0
while b < len(u):
parent_indx = 0
while parent_indx < len(u):
removed = False
S = self.generate_possible_sub_sets_of_size(u, b, u[parent_indx])
test_parent = u[parent_indx]
#print("Test Parent", test_parent)
for parents_set in S:
print("Parent Set", parents_set)
print("Test Parent", test_parent)
if self.complete_test(test_parent, var_id, parents_set, child_states_numb, tot_vars_count):
#print("Removing EDGE:", test_parent, var_id)
self._complete_graph.remove_edge(test_parent, var_id)
u.remove(test_parent)
removed = True
break
if not removed:
parent_indx += 1
b += 1
self._cache.clear()
def generate_possible_sub_sets_of_size(self, u: typing.List, size: int, parent_label: str) -> \
typing.Iterator:
"""
Creates a list containing all possible subsets of the list u of size size,
that do not contains a the node identified by parent_label.
Parameters:
u: the list of _nodes
size: the size of the subsets
parent_label: the _nodes to exclude in the subsets generation
Returns:
a Map Object containing a list of lists
"""
list_without_test_parent = u[:]
list_without_test_parent.remove(parent_label)
return map(list, itertools.combinations(list_without_test_parent, size))
def ctpc_algorithm(self):
"""
Compute the CTPC algorithm.
Parameters:
void
Returns:
void
"""
ctpc_algo = self.one_iteration_of_CTPC_algorithm
total_vars_numb = self._sample_path.total_variables_count
[ctpc_algo(n, total_vars_numb) for n in self._nodes]
def save_results(self):
"""
Save the estimated Structure to a .json file
Parameters:
void
Returns:
void
"""
res = json_graph.node_link_data(self._complete_graph)
name = self._sample_path._importer.file_path.rsplit('/', 1)[-1]
name = 'results_' + name #TODO va aggiunto anche l'indice di array
with open(name, 'w') as f:
json.dump(res, f)