def run_experiment(query_class, function_pipeline, weights, label_coarse):
evaluator = Evaluator(FEATURE_DATA_FILE, label_coarse=label_coarse)
mapping = {
item["label"]: item["label_coarse"]
for item in evaluator.query_matcher.features_raw_init
}
all_data = evaluator.query_matcher.features_raw
my_subset = [
item for item in all_data if item["label"] == (
mapping[query_class] if evaluator.label_coarse else query_class)
]
k_all_db_results_cociwa = evaluator.perform_matching_on_subset_slow(
my_subset, function_pipeline, weights, k_full_db_switch=True)
k_all_db_results_cociwa = evaluator.metric_F1(k_all_db_results_cociwa,
evaluator.mesh_classes_count,
k=10,
use_class_k=True,
silent=True)
k_all_db_results_cociwa[["matches_class"]] = [
Counter(tuple(row[["matches_class"]][0]))
for index, row in k_all_db_results_cociwa.iterrows()
]
results = k_all_db_results_cociwa[["name", "class", "matches_class"]]
results.head(10)
my_result = evaluator.calc_weighted_metric(k_all_db_results_cociwa,
"F1score", "F1")[1]
return my_result
def run_experiment_single(query_mesh, function_pipeline, weights, label_coarse):
evaluator = Evaluator(FEATURE_DATA_FILE, label_coarse=label_coarse)
mapping = {item["label"]: item["label_coarse"] for item in evaluator.query_matcher.features_raw_init}
all_data = evaluator.query_matcher.features_raw
# print(evaluator.mesh_classes_count.keys())
# print(evaluator.mesh_classes_count[evaluator.mesh_classes_count==10])
my_subset = [item for item in all_data if item["name"] == query_mesh]
k_all_db_results_cociwa = evaluator.perform_matching_on_subset_slow(my_subset, function_pipeline, weights, k_full_db_switch=True)
k_all_db_results_cociwa = evaluator.metric_F1(k_all_db_results_cociwa, evaluator.mesh_classes_count, k=10, use_class_k=True)
k_all_db_results_cociwa[["matches_class"]] = [Counter(tuple(row[["matches_class"]][0])) for index, row in k_all_db_results_cociwa.iterrows()]
results = k_all_db_results_cociwa[["name", "class", "matches_class"]]
# print(results.head(10))
print("")
results.head(10)
print(evaluator.calc_weighted_metric(k_all_db_results_cociwa)[1])
class_metric_means = pd.DataFrame(
evaluator.mesh_classes_count.keys().sort_values().to_list(),
columns=['class'])
overall_metrics = []
weights_best_all_features = ([1]) + ([1] * count_hists) + ([1.47] *
count_skeletons)
function_pipeline_cowaeu = [cosine] + ([cityblock] * count_hists) + (
[cityblock] * count_skeletons)
k_all_db_results_cowaeu = evaluator.perform_matching_calculate_metrics(
function_pipeline_cowaeu,
weights_best_all_features,
k_full_db_switch=True)
k_all_db_results_cowaeu.to_json("trash/k_all_db_results_cocici_bw.json")
k_all_db_results_cowaeu = pd.read_json(
"trash/k_all_db_results_cocici_bw.json")
k_all_db_results_cowaeu = evaluator.metric_F1(k_all_db_results_cowaeu,
evaluator.mesh_classes_count,
k=10,
use_class_k=False,
weightbool=True,
weight=0.25)
cowaeu_classes_mean_F1, cowaeu_overall_F1 = evaluator.calc_weighted_metric(
k_all_db_results_cowaeu, "F1score", "F1")
class_metric_means["F1_cowaeu"] = cowaeu_classes_mean_F1
overall_metrics.append(["F1_cowaeu", cowaeu_overall_F1])
evaluator.plot_metric_class_aggregate(
class_metric_means[["class", "F1_cowaeu"]], "F1_cowaeu", "F1Score")
