def get_all_combination_STEVE_best_graph(pred_labels_over_runs, performance_runs): ## steve's suggestions
runs = list(pred_labels_over_runs.keys())
meta_subgraphs = []
for i in range(len(runs)-1):
meta_subgraphs += list(combinations(runs, i+2))
print ("len of meta subgraphs are ", len(meta_subgraphs))
meta_graph_scores = []
for meta_sub_graph1 in meta_subgraphs:
current_subgraph_overlap = []
current_subgraph_perf = []
for ik1, rk1 in enumerate(meta_sub_graph1[:-1]):
current_subgraph_perf.append(performance_runs[rk1])
for rk2 in meta_sub_graph1[ik1+1:]:
current_subgraph_overlap.append (float(calculate_overlap_labels(pred_labels_over_runs[rk1], pred_labels_over_runs[rk2]) ) )
avg_score = sum(current_subgraph_perf)/float(len(current_subgraph_perf))
avg_overlap = sum(current_subgraph_overlap)/float(len(current_subgraph_overlap))
meta_graph_scores.append(avg_score/float(avg_overlap)) ## taking average of subgraph scores
print ("the len of meta graph scores are : ", len(meta_graph_scores))
best_sub_graph_index = meta_graph_scores.index(max(meta_graph_scores))
print ("best subgraph is: ", meta_subgraphs[best_sub_graph_index], max(meta_graph_scores))
return meta_subgraphs[best_sub_graph_index]
def get_complete_graph(pred_labels_over_runs,
performance_runs,
subgraph_size=4):
runs = list(pred_labels_over_runs.keys())
meta_subgraphs = list(combinations(runs, subgraph_size))
print("len of meta subgraphs are ", len(meta_subgraphs))
meta_graph_scores = []
for meta_sub_graph1 in meta_subgraphs:
current_subgraph_score = []
for ik1, rk1 in enumerate(meta_sub_graph1[:-1]):
for rk2 in meta_sub_graph1[ik1 + 1:]:
current_subgraph_score.append(
performance_runs[rk1] * performance_runs[rk2] / float(
calculate_overlap_labels(pred_labels_over_runs[rk1],
pred_labels_over_runs[rk2])))
meta_graph_scores.append(
sum(current_subgraph_score) / float(len(current_subgraph_score)))
print("the len of meta graph scores are : ", len(meta_graph_scores))
best_sub_graph_index = meta_graph_scores.index(max(meta_graph_scores))
print("best subgraph is: ", meta_subgraphs[best_sub_graph_index],
max(meta_graph_scores))
return meta_subgraphs[best_sub_graph_index]
def get_all_combination_best_graph(pred_labels_over_runs,
performance_runs): ## Edge based model
runs = list(pred_labels_over_runs.keys())
meta_subgraphs = []
for i in range(len(runs) - 1):
meta_subgraphs += list(combinations(runs, i + 2))
print("len of meta subgraphs are ", len(meta_subgraphs))
meta_graph_scores = []
for meta_sub_graph1 in meta_subgraphs:
current_subgraph_score = []
for ik1, rk1 in enumerate(meta_sub_graph1[:-1]):
for rk2 in meta_sub_graph1[ik1 + 1:]:
current_subgraph_score.append(
performance_runs[rk1] + performance_runs[rk2] / float(
calculate_overlap_labels(pred_labels_over_runs[rk1],
pred_labels_over_runs[rk2])))
meta_graph_scores.append(
sum(current_subgraph_score) / float(len(current_subgraph_score))
) ## taking average of subgraph scores
print("the len of meta graph scores are : ", len(meta_graph_scores))
best_sub_graph_index = meta_graph_scores.index(max(meta_graph_scores))
print("best subgraph is: ", meta_subgraphs[best_sub_graph_index],
max(meta_graph_scores))
return meta_subgraphs[best_sub_graph_index]
def get_node_pair_score(runs):
for ik1, rk1 in enumerate(runs[:-1]):
for rk2 in runs[ik1 + 1:]:
meta_graph.update({
rk1 + " " + rk2:
calculate_overlap_labels(pred_labels_over_runs[rk1],
pred_labels_over_runs[rk2])
})
def get_all_combination_Vikas_EdgeAPPROACH_withCoverage_best_graph(
pred_labels_over_runs, performance_runs, gold_labels):
runs = list(pred_labels_over_runs.keys())
meta_subgraphs = []
for i in range(len(runs) - 1):
meta_subgraphs += list(combinations(runs, i + 2))
print("len of meta subgraphs are ", len(meta_subgraphs))
meta_graph_scores = []
meta_graph_coverage_scores = []
for meta_sub_graph1 in meta_subgraphs:
current_subgraph_score = []
for ik1, rk1 in enumerate(meta_sub_graph1[:-1]):
if ik1 == 0: ## initializing the coverage list
prediction_coverage = pred_labels_over_runs[rk1]
for rk2 in meta_sub_graph1[ik1 + 1:]:
current_subgraph_score.append(
performance_runs[rk1] + performance_runs[rk2] / float(
calculate_overlap_labels(pred_labels_over_runs[rk1],
pred_labels_over_runs[rk2])))
prediction_coverage = get_union(prediction_coverage,
pred_labels_over_runs[rk2])
final_coverage = sum(get_intersection(
prediction_coverage, gold_labels)) / float(sum(gold_labels))
meta_graph_coverage_scores.append(final_coverage)
meta_graph_scores.append(
(sum(current_subgraph_score) / float(len(current_subgraph_score)))
* final_coverage) ## taking average of subgraph scores
print("the len of meta graph scores are : ", len(meta_graph_scores))
best_sub_graph_index = meta_graph_scores.index(max(meta_graph_scores))
print("best subgraph is: ", meta_subgraphs[best_sub_graph_index],
max(meta_graph_scores),
meta_graph_coverage_scores[best_sub_graph_index])
return meta_subgraphs[best_sub_graph_index]