def preprocess_before_running_model(source_file_, facts_file_,
confidence_value_computation_info_dir_,
dataitem_index_file_, g_):
# load source information
header = False # dictionary with the original trustworthiness
T_actual_ = utils_dataset.load_sources_info(source_file_, header)
T_ = utils_dataset.load_sources_info(source_file_, header)
print(str(len(T_)) + " sources loaded")
# load fact information
header = True
sources_dataItemValues_ = utils_dataset.load_facts(facts_file_, header)
# load data item set
D_ = list(sources_dataItemValues_.keys())
# compute (1) all the facts that are claimed by a source and (2) all the sources that claim a specific fact
# (1) set of facts that are claimed by a specific source < key = source id, value = set of facts (dataitem + value) >
# (2) all the sources that claim a specific fact <key = dataitem + value, value = set of source ids>
print("Fact loading")
fact_and_source_info_ = utils_dataset.load_fact_and_source_info(
sources_dataItemValues_)
F_s_ = fact_and_source_info_[0]
S_ = fact_and_source_info_[1]
print("Computing sources for " + str(len(sources_dataItemValues_)) +
" data items FOR COMPUTATION PURPOSE")
if not (len(os.listdir(confidence_value_computation_info_dir_))
== len(D_)):
# compute the files for belief propagation information
print("graph nodes " + str(len(g_.nodes)))
print("LENGH source data item values" +
str(len(sources_dataItemValues_.values())))
res = utils_taxonomy.create_value_info_computation(
g_, sources_dataItemValues_, dataitem_index_file_,
confidence_value_computation_info_dir_)
sources_dataItemValues_.clear()
header = True
sources_dataItemValues_ = utils_dataset.load_facts(facts_file_, header)
if res:
print("Computation DONE")
# else: the files for contained the info for the belief propagation have been already computed
# then load the relative dataitem id for using the files
dataitem_ids_ = utils_dataset.load_dataitem_ids(dataitem_index_file_)
# load the information
dataitem_values_info_ = utils_dataset.load_all_dataitem_values_confidence_infos_low_memory(
dataitem_ids_, confidence_value_computation_info_dir_,
sources_dataItemValues_)
# S_prop is a dictionary contained for each fact all the sources that it has to take into account for leveraging the belief propagation framework
S_prop_ = dataitem_values_info_[2]
app_conf_dict_ = dataitem_values_info_[3]
app_source_dict_ = dataitem_values_info_[4]
# delete folder confidence_value_computation_info_dir_ and the relative index file
#shutil.rmtree(confidence_value_computation_info_dir_)
# os.remove(dataitem_index_file_)
return [
T_, T_actual_, sources_dataItemValues_, D_, F_s_, S_, S_prop_,
app_conf_dict_, app_source_dict_
]
def preprocess_before_running_model_only_trad(source_file_, facts_file_):
# load source information
header = False # dictionary with the original trustworthiness
T_ = utils_dataset.load_sources_info(source_file_, header)
print(str(len(T_)) + " sources loaded")
# load fact information
header = True
sources_dataItemValues_ = utils_dataset.load_facts(facts_file_, header)
# load data item set
# compute (1) all the facts that are claimed by a source and (2) all the sources that claim a specific fact
# (1) set of facts that are claimed by a specific source < key = source id, value = set of facts (dataitem + value) >
# (2) all the sources that claim a specific fact <key = dataitem + value, value = set of source ids>
print("Fact loading")
fact_and_source_info_ = utils_dataset.load_fact_and_source_info(
sources_dataItemValues_)
F_s_ = fact_and_source_info_[0]
S_ = fact_and_source_info_[1]
sources_dataItemValues_.clear()
header = True
sources_dataItemValues_ = utils_dataset.load_facts(facts_file_, header)
return [T_, sources_dataItemValues_, F_s_, S_]
def analysis_trustworthiness_estimations(dataset_dir, dir_path,
output_error_rate_dir):
if not os.path.exists(output_error_rate_dir):
os.makedirs(output_error_rate_dir)
for root, dirs, files in os.walk(dir_path):
if ["EXP", "LOW_E", "UNI"] == dirs: continue
if ["EXP"] == dirs: continue
if ["UNI"] == dirs: continue
if ["LOW_E"] == dirs: continue
if root.count('/') != 3: continue
for dir_name in dirs:
#print("dir name :" + str(dir_name))
dir_name = dir_name.replace("dataset", "")
n_dataset = dir_name
if "-" in n_dataset:
n_dataset = n_dataset.replace("UNI-", "")
n_dataset = n_dataset.replace("EXP-", "")
n_dataset = n_dataset.replace("LOW_E-", "")
n_dataset = n_dataset.replace("-", "")
n_folder_app = "-" + str(n_dataset[0:2])
n_dataset = n_dataset[2:]
n_folder_app = n_folder_app + n_dataset
else:
n_dataset = n_dataset.replace("UNI_", "")
n_dataset = n_dataset.replace("EXP_", "")
n_dataset = n_dataset.replace("LOW_E_", "")
n_dataset = n_dataset.replace("_", "")
n_folder_app = "_"
n_folder_app = n_folder_app + n_dataset
# facts file path
if "UNI" in root:
subfolder_path_results = "UNI/"
subfolder_path = "UNI/dataset" + str(n_folder_app) + "/"
source_file = dataset_dir + "/UNI/dataset" + str(
n_folder_app) + "/Output_acc_" + str(n_dataset) + ".txt"
if "EXP" in root:
subfolder_path_results = "EXP/"
subfolder_path = "EXP/dataset" + str(n_folder_app) + "/"
source_file = dataset_dir + "/EXP/dataset" + str(
n_folder_app) + "/Output_acc_" + str(n_dataset) + ".txt"
if "LOW_E" in root:
subfolder_path_results = "LOW_E/"
subfolder_path = "LOW_E/dataset" + str(n_folder_app) + "/"
source_file = dataset_dir + "/LOW_E/dataset" + str(
n_folder_app) + "/Output_acc_" + str(n_dataset) + ".txt"
output_estimations = os.path.join(output_estimations_main_dir,
subfolder_path)
output_error_rate = os.path.join(output_error_rate_dir,
subfolder_path_results)
if not os.path.exists(output_error_rate):
os.makedirs(output_error_rate)
output_file_path = output_error_rate + "/trust_error_rate_dataset_" + str(
n_folder_app) + ".csv"
header = False # dictionary with the original trustworthiness
T_actual = utils_dataset.load_sources_info(source_file, header)
trust_file_trad = os.path.join(
output_estimations,
"trad_est_trust_" + str(n_dataset) + ".csv")
trust_trad = utils_dataset.read_trust_estimation_file(
trust_file_trad)
error_rate_trad = compute_error_rate_for_trustworthiness(
T_actual, trust_trad)
# adapted model 1 (only confidence) results files and dir
trust_file_adapt = os.path.join(
output_estimations,
"adapt_est_trust_" + str(n_dataset) + ".csv")
trust_adapt = utils_dataset.read_trust_estimation_file(
trust_file_adapt)
error_rate_adapt = compute_error_rate_for_trustworthiness(
T_actual, trust_adapt)
utils_writing_results.writing_trustworthiness_error_rate_file(
output_file_path, error_rate_trad, error_rate_adapt)
return True
if AdaptedSums_and_Rules_flag:
res_list = preprocessing_sums_model.preprocess_before_running_model(source_file, facts_file,
confidence_value_computation_info_dir,
dataitem_index_file,
g)
T = res_list[0]
T_actual = res_list[1]
sources_dataItemValues = res_list[2]
D = res_list[3]
F_s = res_list[4]
S = res_list[5]
S_prop = res_list[6]
app_conf_dict = res_list[7]
header = False # original trustworthiness file
T_actual = utils_dataset.load_sources_info(source_file, header)
header = False # load trustworthiness of sources
T = utils_dataset.load_sources_info(source_file, header)
print(str(len(T)) + " sources loaded")
S_set = list(T.keys())
# load facts
header = True
sources_dataItemValues = utils_dataset.load_facts(facts_file, header)
D = list(sources_dataItemValues.keys())
# compute (1) all the facts that are claimed by a source and (2) all the sources that claim a specific fact
# (1) set of facts that are claimed by a specific source < key = source id, value = set of facts (dataitem + value) >
# (2) all the sources that claim a specific fact <key = dataitem + value, value = set of source ids>
print("Fact loading")