/* * Eliminates all variables defined as evidence. * The order of the variables that are not eliminated is * the same order in the original function. */ private ProbabilityFunction check_evidence(ProbabilityFunction pf) { int i, j, k, v, aux_i; boolean markers[] = new boolean[bn.number_variables()]; int n = build_evidence_markers(pf, markers); // Handle special cases if (n == 0) return (null); // No variable remains if (n == pf.number_variables()) return (pf); // No relevant evidence // Calculate necessary quantities in such a // way that the order of variables in the original // function is not altered. int joined_indexes[] = new int[n]; for (i = 0, j = 0, v = 1; i < pf.number_variables(); i++) { aux_i = pf.get_variable(i).get_index(); if (markers[aux_i] == true) { joined_indexes[j] = aux_i; j++; v *= bn.get_probability_variable(aux_i).number_values(); } } // Create new function to be filled with joined variables ProbabilityFunction new_pf = new ProbabilityFunction(bn, n, v, null); for (i = 0; i < n; i++) new_pf.set_variable(i, bn.get_probability_variable(joined_indexes[i])); // Loop through the values check_evidence_loop(new_pf, pf); return (new_pf); }
/* * Transform an observed ProbabilityVariable into a ProbabilityFunction * to handle the case where the query involves an observed variable. */ private ProbabilityFunction transform_to_probability_function( BayesNet bn, ProbabilityVariable pv) { ProbabilityFunction pf = new ProbabilityFunction(bn, 1, pv.number_values(), null); pf.set_variable(0, pv); int index_of_value = pv.get_observed_index(); pf.set_value(index_of_value, 1.0); return (pf); }
/** * Constructor for BucketTree. Does the whole initialization; it should be the only method that * deals with symbolic names for variables. */ public BucketTree(Ordering ord, boolean dpc) { int i, j, markers[]; ProbabilityFunction pf; ProbabilityVariable pv; DiscreteVariable aux_pv; DiscreteFunction ut; String order[]; do_produce_clusters = dpc; ordering = ord; // Collect information from the Ordering object. bn = ord.bn; explanation_status = ord.explanation_status; order = ord.order; // Indicate the first bucket to process active_bucket = 0; // Check the possibility that the query has an observed variable i = bn.index_of_variable(order[order.length - 1]); pv = bn.get_probability_variable(i); if (pv.is_observed() == true) { pf = transform_to_probability_function(bn, pv); bucket_tree = new Bucket[1]; bucket_tree[0] = new Bucket(this, pv, do_produce_clusters); insert(pf); } else { // Initialize the bucket objects bucket_tree = new Bucket[order.length]; for (i = 0; i < order.length; i++) { j = bn.index_of_variable(order[i]); bucket_tree[i] = new Bucket(this, bn.get_probability_variable(j), do_produce_clusters); } // Insert the probability functions into the bucket_tree; // first mark all functions that are actually going // into the bucket_tree. markers = new int[bn.number_variables()]; for (i = 0; i < order.length; i++) markers[bn.index_of_variable(order[i])] = 1; // Now insert functions that are marked and non-null. for (i = 0; i < bn.number_probability_functions(); i++) { if (markers[bn.get_probability_function(i).get_index(0)] == 1) { pf = check_evidence(bn.get_probability_function(i)); if (pf != null) { aux_pv = (bn.get_probability_function(i)).get_variable(0); insert(pf, !pf.memberOf(aux_pv.get_index())); } } } // Insert the utility_function. ut = bn.get_utility_function(); if (ut != null) insert(ut); } }
/* * Obtain the values for the evidence plus function. */ private void check_evidence_loop(ProbabilityFunction new_pf, ProbabilityFunction pf) { int i, j, k, l, m, p, last, current; int indexes[] = new int[bn.number_variables()]; int value_lengths[] = new int[bn.number_variables()]; for (i = 0; i < bn.number_variables(); i++) { indexes[i] = 0; value_lengths[i] = bn.get_probability_variable(i).number_values(); } for (i = 0; i < bn.number_variables(); i++) { if (bn.get_probability_variable(i).is_observed()) { indexes[i] = bn.get_probability_variable(i).get_observed_index(); } } last = new_pf.number_variables() - 1; for (i = 0; i < new_pf.number_values(); i++) { p = new_pf.get_position_from_indexes(indexes); new_pf.set_value(p, pf.evaluate(indexes)); indexes[new_pf.get_index(last)]++; for (j = last; j > 0; j--) { current = new_pf.get_index(j); if (indexes[current] >= value_lengths[current]) { indexes[current] = 0; indexes[new_pf.get_index(j - 1)]++; } else break; } } }
/* * Build an array of markers. The marker for a * variable is true only if the variable is present in the * input ProbabilityFunction pf and is not observed. * Even explanatory variables can be observed and taken as * evidence. */ private int build_evidence_markers(ProbabilityFunction pf, boolean markers[]) { int i, n; // Initialize the markers for (i = 0; i < markers.length; i++) markers[i] = false; // Insert the variables of the ProbabilityFunction for (i = 0; i < pf.number_variables(); i++) markers[pf.get_index(i)] = true; // Take the evidence out for (i = 0; i < bn.number_variables(); i++) { if (bn.get_probability_variable(i).is_observed()) markers[i] = false; } // Count how many variables remain n = 0; for (i = 0; i < markers.length; i++) { if (markers[i] == true) n++; } return (n); }
/** Get the normalized result for the BucketTree. */ public ProbabilityFunction get_normalized_result() { ProbabilityFunction aux_pf = new ProbabilityFunction(unnormalized_result, bn); aux_pf.normalize(); return (aux_pf); }
/** Set a probability variable given its index. */ public void set_probability_function(int index, ProbabilityFunction p_f) { p_f.bn = this; probability_functions[index] = p_f; }