def gather_statistics(schema_set, datasources_with_tag, stats_output_p, datasets_input_p):
for source_name in datasources_with_tag:
all_stats = {}
file_path = stats_output_p + source_name + '.json'
my_file = pathlib.Path(file_path)
if my_file.exists():
continue
path = datasets_input_p + source_name + '.csv'
dataset = pd.read_csv(path, index_col=0, header=0)
dataset = df_rename_cols(dataset)
attr_schema = schema_set[source_name]
df_columns = list(dataset.columns.values)
# print(attr_schema)
attr_schema = [{'name': parse_dataset.clean_name(attr['name'], False, False)} for attr in attr_schema]
df_columns = [parse_dataset.clean_name(attr, False, False) for attr in df_columns]
cols_to_delete = find_attrs_to_delete(attr_schema, df_columns)
dataset = df_delete_cols(dataset, cols_to_delete)
schema = schema_set[source_name]
attributes_list = [parse_dataset.clean_name(attr['name'], False, False) for attr in schema]
attributes_list = [item for item in attributes_list if item not in cols_to_delete]
print(source_name, attributes_list, cols_to_delete)
for attr in attributes_list:
stat, groups, uniques = groupby_unique(attr, dataset)
all_stats[attr] = stat
# TODO more types of stats needed
with open(file_path, 'w') as fp:
json.dump(all_stats, fp, sort_keys=True, indent=2)
return
def load_prematching_metadata(p, m, pds):
all_topics = {}
topic_contexts_f = open(p.enriched_topics_json_dir, 'r')
topic_contexts = json.load(topic_contexts_f)
all_topics = {}
# TODO get all available topics
for source_name in m.datasources_with_tag:
source_name = pds.clean_name(source_name, False, False)
for src_top in topic_contexts[source_name]:
if src_top not in all_topics:
all_topics[src_top] = {}
all_topics[src_top][source_name] = topic_contexts[source_name][
src_top] # synsets
attrs_contexts_f = open(p.enriched_attrs_json_dir, 'r')
attrs_contexts = json.load(attrs_contexts_f)
return all_topics, attrs_contexts, topic_contexts
def load_per_source_metadata(p, m, datasources, source_name, pds, bmm):
# path = datasets_path + source_name + '.csv'
# dataset = pd.read_csv(path, index_col=0, header=0)
stats_f = open(p.dataset_stats + source_name + '.json', 'r')
stats = json.load(stats_f)
df_columns = list(stats.keys())
schema = m.schema_set[source_name]
metadata = m.dataset_metadata_set[source_name]['tags']
dataset = pd.DataFrame()
# dataset = bmm.df_rename_cols(dataset)
datasources[source_name] = (source_name, dataset, schema, metadata)
print(source_name)
# bmm.print_metadata_head(source_name, dataset, schema, metadata)
# initialization schema matching
tags_list = [tag['display_name'] for tag in metadata]
# use enriched tags instead
tags_list_enriched_f = open(p.enriched_topics_json_dir, 'r')
tags_list_enriched = json.load(tags_list_enriched_f)
tags_list_enriched_dataset = tags_list_enriched[source_name]
tags_list_enriched_names = list(tags_list_enriched[source_name].keys())
# TODO add non-overlapping homonyms to context
attributes_list = [
pds.clean_name(attr['name'], False, False) for attr in schema
]
cols_to_delete = bmm.find_attrs_to_delete(attributes_list, df_columns)
attributes_list = [
item for item in attributes_list if item not in cols_to_delete
]
return tags_list_enriched_dataset, tags_list_enriched_names, attributes_list, schema, datasources, stats
def print_metadata_head(source_name, dataset, schema, metadata):
print('dataset_name:', source_name)
print('dataset_values.head \n', dataset.head())
print('dataset_schema[0]', parse_dataset.clean_name(schema[0]['name'], False, False))
print('dataset_schema[1]', parse_dataset.clean_name(schema[1]['name'], False, False))
print('dataset_tags[0]', metadata[0]['display_name'])
def df_rename_cols(dataset):
dataset_col_names = [parse_dataset.clean_name(x) for x in list(dataset.columns.values)]
col_rename_dict = {i: j for i, j in zip(list(dataset.columns.values), dataset_col_names)}
dataset.rename(columns=col_rename_dict, inplace=True)
return dataset
datasets_path = './thesis_project_dataset_clean/'
datasources = {}
for source_name in datasources_with_tag:
dataset = pd.read_csv(datasets_path + source_name + '.csv', index_col=0, header=0)
schema = schema_set[source_name]
metadata = dataset_metadata_set[source_name]['tags']
dataset = df_rename_cols(dataset)
datasources[source_name] = (source_name, dataset, schema, metadata)
print_metadata_head(source_name, dataset, schema, metadata)
# initialization schema matching
tags_list = [tag['display_name'] for tag in metadata]
attributes_list = [parse_dataset.clean_name(attr['name'], False, False) for attr in schema]
sim_matrix = build_local_similarity_matrix(tags_list, attributes_list)
sim_frame = pd.DataFrame(data=sim_matrix, columns=attributes_list, index=tags_list)
print(sim_frame.to_string())
tree_matches = sim_frame.loc['trees']
attrs = list(sim_frame.columns.values)
max_score = 0
arg_max_score = None
arg_i = -1
for attr_i in range(len(attrs)):
attr = attrs[attr_i]
score = sim_frame.loc['trees', attr]
if score > max_score:
max_score = score
arg_max_score = attr
def perform_matching(p, dataset_metadata_set, schema_set, datasources_with_tag,
kb, params):
comparison_count = 0
comparison_count_o = [comparison_count]
sim_matrices = {}
for source_name in datasources_with_tag:
t2 = time.time()
dataset = pd.read_csv(p.datasets_path + source_name + '.csv',
index_col=0,
header=0)
dataset = bmm.df_rename_cols(dataset)
schema = schema_set[source_name]
metadata = dataset_metadata_set[source_name]['tags']
schema = [{
'name': pds.clean_name(attr['name'], False, False)
} for attr in schema]
schema_attr_names = []
for attr in schema:
# attr['name'] = pds.clean_name(attr['name'], False, False)
schema_attr_names.append(attr['name'])
schema_attr_names.sort()
for concept in kb:
for datasource in kb[concept]['matches']:
src_attr = kb[concept]['matches'][datasource]['attribute']
src_vals = kb[concept]['matches'][datasource]['example_values']
# do not match with self
if source_name == datasource:
continue
# do not match if no populated values
if src_vals == None:
continue
src_data = pd.DataFrame({src_attr: src_vals})
print("[concept:%s, datasource:%s(%s) <=> dataset:%s]" %
(concept, datasource, src_attr, source_name))
# groupby values for each column and obtain count for each unique value, then multiply counts when comparison succeeds
tar_schema = list(dataset.columns.values)
cols_to_delete = bmm.find_attrs_to_delete(schema, tar_schema)
tar_schema = [
item for item in tar_schema if item not in cols_to_delete
]
attrs_stat = {}
max_len = 0
for attr in tar_schema:
# TODO save this output to file for later use
# stat, groups, uniques = bmm.groupby_unique(attr, dataset)
stat, uniques = bmm.get_attr_stats(p.dataset_stats,
source_name, attr)
uniques.sort()
# save for later
try:
arg_i = schema_attr_names.index(attr)
if schema[arg_i]['domain'] == None:
schema[arg_i]['coded_values'] = uniques
schema[arg_i]['domain'] = 'coded_values_groupby'
except:
pass
attrs_stat[attr] = (stat, uniques)
if len(uniques) > max_len:
max_len = len(uniques)
tar_df = pd.DataFrame()
for attr in tar_schema:
uniques = attrs_stat[attr][1]
attrs_stat[attr] = attrs_stat[attr][0]
attr_vals = uniques + ['None'] * (max_len - len(uniques))
tar_df[attr] = attr_vals
# collect stats first, also compare data types
src_datatype = kb[concept]['matches'][datasource]['data_type']
attr_schema = schema_set[datasource]
cols_to_delete = bmm.compare_datatypes(src_datatype,
attr_schema, tar_schema)
tar_df = bmm.df_delete_cols(tar_df, cols_to_delete)
# TODO datatypes must match! need to move to a different matcher
sim_matrix, confidence = bmm.match_table_by_values(
src_data, tar_df, params['match_threshold'],
comparison_count_o, attrs_stat, params['sample_ratio'],
params['sample_min_count'], params['sample_max_count'])
src_names = list(src_data.columns.values)
tar_names = list(tar_df.columns.values)
sim_matrix2 = schm.matcher_name_matrix(src_names, tar_names)
sim_matrix = schm.combine_scores_matrix(
sim_matrix, sim_matrix2, params['proportions'])
print(sim_matrix.to_string())
# save similarity matrices
filename = '%s|%s|%s||%s.csv' % (concept, datasource, src_attr,
source_name)
sim_matrices[filename] = sim_matrix
t3 = time.time()
total = t3 - t2
print('time %s sec' % (total))
print('-----')
return kb, sim_matrices
def initialize_matching(p, input_topics, dataset_metadata_set, schema_set,
datasources_with_tag, reverse_index, kb):
datasources = {}
for source_name in datasources_with_tag:
# path = datasets_path + source_name + '.csv'
# dataset = pd.read_csv(path, index_col=0, header=0)
stats_f = open(p.dataset_stats + source_name + '.json', 'r')
stats = json.load(stats_f)
df_columns = list(stats.keys())
schema = schema_set[source_name]
metadata = dataset_metadata_set[source_name]['tags']
dataset = pd.DataFrame()
# dataset = bmm.df_rename_cols(dataset)
datasources[source_name] = (source_name, dataset, schema, metadata)
print(source_name)
if DEBUG_MODE:
bmm.print_metadata_head(source_name, dataset, schema, metadata)
# initialization schema matching
tags_list = [tag['display_name'] for tag in metadata]
attributes_list = [
pds.clean_name(attr['name'], False, False) for attr in schema
]
cols_to_delete = bmm.find_attrs_to_delete(attributes_list, df_columns)
attributes_list = [
item for item in attributes_list if item not in cols_to_delete
]
sim_matrix = bmm.build_local_similarity_matrix(tags_list,
attributes_list)
sim_frame = pd.DataFrame(data=sim_matrix,
columns=attributes_list,
index=tags_list)
# print(sim_frame.to_string())
# TODO during new concepts stage, add second best tag and so on
attrs = list(sim_frame.columns.values)
# if stats file is empty
if len(attrs) == 0:
return kb, datasources_with_tag, schema_set
for topic in reverse_index[source_name]:
max_score = 0
arg_max_score = None
arg_i = -1
for attr_i in range(len(attrs)):
attr = attrs[attr_i]
score = sim_frame.loc[topic, attr]
if score > max_score:
max_score = score
arg_max_score = attr
arg_i = attr_i
arg_max_examples_vals = None
example_value = None
if schema[arg_i]['domain'] != None:
arg_max_examples_vals = schema[arg_i]['coded_values']
arg_max_examples_vals.sort()
if len(arg_max_examples_vals) > 0:
example_value = arg_max_examples_vals[0]
else:
# loading from stats file
_, uniques = bmm.get_attr_stats(p.dataset_stats, source_name,
attrs[arg_i])
# stat, _, uniques = bmm.groupby_unique(attrs[arg_i], dataset)
uniques.sort()
schema[arg_i]['coded_values'] = uniques
arg_max_examples_vals = schema[arg_i]['coded_values']
if len(arg_max_examples_vals) > 0:
print('arg_max_examples_vals', arg_max_examples_vals[0])
schema[arg_i]['domain'] = 'coded_values_groupby'
print('best match:', topic, arg_max_score, max_score,
example_value)
kb_match_entry = {
'concept': topic,
'datasource': source_name,
'attribute': arg_max_score,
'match_score': max_score,
'example_values': arg_max_examples_vals,
'data_type': schema[arg_i]['data_type']
}
bmm.update_kb_json(kb, kb_match_entry)
print('-----')
# done initialization
return kb, datasources_with_tag, schema_set
def create_attributes_contexts(datasets, m, p, r):
contexts = {}
for dataset in datasets:
contexts[dataset] = {}
schema = m.schema_set[dataset]
attributes_list = [
pds.clean_name(attr['name'], False, False) for attr in schema
]
dataset_existing_tags = m.dataset_metadata_set[dataset]['tags']
dataset_existing_groups = m.dataset_metadata_set[dataset]['groups']
dataset_notes = m.dataset_metadata_set[dataset]['notes']
desc = ''
for group in dataset_existing_groups:
desc = ' ' + group['description']
dataset_existing_tags = [
tag['display_name'] for tag in dataset_existing_tags
]
dataset_existing_groups = [
group['display_name'] for group in dataset_existing_groups
]
dataset_notes = [
word for word in dataset_notes.split() if "http://" not in word
]
notes = ' '.join(dataset_notes)
stats_f = open(p.dataset_stats + dataset + '.json', 'r')
stats = json.load(stats_f)
df_columns = list(stats.keys())
attributes_list = [
pds.clean_name(attr['name'], False, False) for attr in schema
]
cols_to_delete = bmm.find_attrs_to_delete(attributes_list, df_columns)
attributes_list = [
item for item in attributes_list if item not in cols_to_delete
]
for attr in attributes_list:
# other_attrs = attributes_list.copy()
# other_attrs.remove(attr)
other_attrs = []
attr_values = stats[attr].keys()
# TODO get average of val length, place attr vals in notes if length is long
length = 0
if len(attr_values) > 0:
if len(attr_values) > r.vals_truncate_sample:
num_to_select = r.vals_truncate_sample
attr_values = random.sample(attr_values, num_to_select)
length = [len(val) for val in attr_values]
length = sum(length) / len(attr_values)
if r.sentence_threshold <= length:
notes = notes + '. ' + '. '.join([val for val in attr_values])
# print('>>>>>', notes)
else:
other_attrs.extend(attr_values)
# print('>>>>>', other_attrs)
pt.enrich_homonyms(dataset, attr, desc, notes, other_attrs)
m.dataset_attributes_contexts = contexts
return contexts
def initialize_matching(p, m, r):
all_topics, attrs_contexts, topic_contexts = load_prematching_metadata(
p, m, pds)
wordnet = pt.load_dict()
pair_dict_all = {}
datasources = {}
for source_name in m.datasources_with_tag:
tags_list_enriched_dataset, tags_list_enriched_names, attributes_list, schema, _, _ = load_per_source_metadata(
p, m, datasources, source_name, pds, bmm)
score_names = m.score_names
sim_matrix1 = build_local_similarity_matrix(tags_list_enriched_dataset,
attributes_list, r)
# TODO build_local_similarity_matrix using context
if source_name not in attrs_contexts:
print('ERROR: DATASOURCE NOT FOUND', source_name, '\n', '-----')
continue
attribute_contexts = attrs_contexts[source_name]
# topic_contexts is all datasets, attribute_contexts is per dataset
sim_matrix2, sim_matrix3, pair_dict = build_local_context_similarity_matrix(
topic_contexts, attribute_contexts, source_name, wordnet,
all_topics)
sim_frame1 = pd.DataFrame(data=sim_matrix1,
columns=attributes_list,
index=tags_list_enriched_names)
sim_frame2 = pd.DataFrame(data=sim_matrix2,
columns=attributes_list,
index=tags_list_enriched_names)
sim_frame3 = pd.DataFrame(data=sim_matrix3,
columns=attributes_list,
index=tags_list_enriched_names)
# chance of getting external topics
pair_dict_all.update(pair_dict)
# print(sim_frame.to_string())
attrs = list(sim_frame1.columns.values)
# if stats file is empty
if len(attrs) == 0:
print('ERROR: empty dataset', source_name, '\n', '-----')
continue
# get example values
for attr_i in range(len(schema)):
# print(attr_i)
if 'domain' not in schema[attr_i] or schema[attr_i][
'domain'] == None:
attr_name = schema[attr_i]['name']
attr_name = pds.clean_name(attr_name, False, False)
# loading from stats file
_, uniques = bmm.get_attr_stats(p.dataset_stats, source_name,
attr_name)
if uniques != None:
# print('uniques', len(uniques))
pass
else:
continue
# stat, _, uniques = bmm.groupby_unique(attrs[arg_i], dataset)
uniques.sort()
schema[attr_i]['coded_values'] = uniques
# arg_max_examples_vals = schema[attr_i]['coded_values']
# if len(arg_max_examples_vals) > 0: print('arg_max_examples_vals', arg_max_examples_vals[0])
schema[attr_i]['domain'] = 'coded_values_groupby'
# init kb
build_kb_json(tags_list_enriched_names, source_name, m)
# during new concepts stage, add second best tag and so on
for attr_i in range(len(attrs)):
scores1 = [[
attr_i, attrs[attr_i], sim_frame1.loc[topic, attrs[attr_i]],
topic
] for topic in tags_list_enriched_names]
scores2 = [[
attr_i, attrs[attr_i], sim_frame2.loc[topic, attrs[attr_i]],
topic
] for topic in tags_list_enriched_names]
scores3 = [[
attr_i, attrs[attr_i], sim_frame3.loc[topic, attrs[attr_i]],
topic
] for topic in tags_list_enriched_names]
score_len = 0
if len(scores1) != 0: score_len = len(scores1[0])
# for topic in tags_list_enriched_names:
# scores1 = [[attr_i, attrs[attr_i], sim_frame1.loc[topic, attrs[attr_i]]] for attr_i in range(len(attrs))]
# scores2 = [[attr_i, attrs[attr_i], sim_frame2.loc[topic, attrs[attr_i]]] for attr_i in range(len(attrs))]
scores = []
for i in range(len(scores1)):
if scores1[i][2] >= scores2[i][2]:
# print(scores2[i][2])
# print(scores3[i][2])
scores.append([
attr_i, attrs[attr_i], scores1[i][2], scores1[i][3],
score_names[0]
])
else:
multiplier = 1.0
if scores3[i][2] != 0: multiplier = scores3[i][2]
scores.append([
attr_i, attrs[attr_i],
min(scores2[i][2] * multiplier, 1.0), scores2[i][3],
score_names[1]
])
scores = sorted(scores, key=lambda tup: tup[2])
scores.reverse()
scores_examples = []
for attr_score in scores:
# example_value = None
# print(attr_score, attr_score[0], schema[attr_score[0]])
if 'coded_values' not in schema[attr_score[0]]:
continue
arg_max_examples_vals = schema[attr_score[0]]['coded_values']
arg_max_examples_vals.sort()
scores_examples.append(attr_score +
[schema[attr_score[0]]['coded_values']])
# print('here')
top = 0
output = []
for score in scores_examples:
if len(score) <= score_len:
# print('skip', score)
continue
# print('topic_to_attr_count', score[2], top)
if score[
2] > r.topic_to_attr_threshold and top <= r.topic_to_attr_count:
# print('topic_to_attr_count', r.topic_to_attr_count)
output.append(score)
top += 1
# if len(output) == 0:
# output.append(scores_examples[0])
# max_score = 0
# arg_max_score = None
# arg_i = -1
# for attr_i in range(len(attrs)):
# attr = attrs[attr_i]
# score = sim_frame.loc[topic, attr]
# if score > max_score:
# max_score = score
# arg_max_score = attr
# arg_i = attr_i
# if len(arg_max_examples_vals) > 0: example_value = arg_max_examples_vals[0]
# print('best match:', topic, arg_max_score, max_score, example_value)
# print('=====output', output)
for match in output:
kb_match_entry = {
'concept': match[3],
'datasource': source_name,
'attribute': match[1],
'match_score': match[2],
'example_values': match[5],
'data_type': schema[match[0]]['data_type'],
'score_name': match[4]
}
update_kb_json(m.kbs[source_name], kb_match_entry)
# for debugging:
kb_match_entry['example_values'] = kb_match_entry[
'example_values'][:min(
len(kb_match_entry['example_values']), 5)]
pprint.pprint(kb_match_entry)
print('-----')
m.pair_dict_all = pair_dict_all
# done initialization
return True
def initialize_matching_full(p, m, r):
all_topics, attrs_contexts, topic_contexts = load_prematching_metadata(
p, m, pds)
wordnet = pt.load_dict()
pair_dict_all = {}
kb_curr_file = './outputs/kb_file.json' # TODO <=
if not os.path.exists(kb_curr_file):
with open(kb_curr_file, 'w') as fp:
json.dump({}, fp, sort_keys=True, indent=2)
fp = open(kb_curr_file, 'r')
m.kbs = json.load(fp)
len_all_ds = len(m.datasources_with_tag)
datasources = {}
for source_name in m.datasources_with_tag:
if source_name in m.kbs:
print('--already have local mapping ', source_name)
continue # already have local mapping, skip
_, _, attributes_list, schema, _, _ = load_per_source_metadata(
p, m, datasources, source_name, pds, bmm)
attributes_list_orig = [
pds.clean_name(attr['name'], False, False) for attr in schema
]
score_names = m.score_names
# 700+ topics
sim_matrix1 = build_local_similarity_matrix(all_topics,
attributes_list, r)
if source_name not in attrs_contexts:
print('ERROR: DATASOURCE NOT FOUND', source_name, '\n', '---!--')
continue
attribute_contexts = attrs_contexts[source_name]
# topic_contexts is all datasets, attribute_contexts is per dataset
sim_matrix2, pair_dict, sim_matrix3 = build_local_context_similarity_matrix_full(
topic_contexts, attribute_contexts, source_name, wordnet,
all_topics, p.server_ip)
pprint.pprint(pair_dict)
sim_frame1 = pd.DataFrame(data=sim_matrix1,
columns=attributes_list,
index=all_topics.keys())
sim_frame2 = pd.DataFrame(data=sim_matrix2,
columns=list(attribute_contexts.keys()),
index=all_topics.keys())
sim_frame3 = pd.DataFrame(data=sim_matrix3,
columns=list(attribute_contexts.keys()),
index=all_topics.keys())
pair_dict_all.update(pair_dict)
attrs = list(sim_frame1.columns.values)
if len(attrs) == 0:
print('ERROR: empty dataset', source_name, '\n', '-----')
continue
for attr_i in range(len(schema)):
if 'domain' not in schema[attr_i] or schema[attr_i][
'domain'] == None:
attr_name = schema[attr_i]['name']
attr_name = pds.clean_name(attr_name, False, False)
_, uniques = bmm.get_attr_stats(p.dataset_stats, source_name,
attr_name)
if uniques != None:
pass
else:
continue
uniques.sort()
schema[attr_i]['coded_values'] = uniques
schema[attr_i]['domain'] = 'coded_values_groupby'
# init kb
build_kb_json(all_topics, source_name, m)
for attr_i in range(len(attrs)):
scores1 = [[
attr_i, attrs[attr_i], sim_frame1.loc[topic, attrs[attr_i]],
topic, None
] for topic in all_topics]
scores2 = [[
attr_i, attrs[attr_i], sim_frame2.loc[topic,
attrs[attr_i]], topic,
pair_dict[(source_name, pds.clean_name(attrs[attr_i]), topic)]
] for topic in all_topics]
scores3 = [[
attr_i, attrs[attr_i], sim_frame3.loc[topic, attrs[attr_i]],
topic, None
] for topic in all_topics]
score_len = 0
if len(scores1) != 0: score_len = len(scores1[0])
# TODO change score to weighted average
weights = [0.40, 0.6, 0.0] # TODO train the weights
scores = []
for i in range(len(scores1)):
scores_tmp = [scores1[i][2], scores2[i][2], scores3[i][2]]
index, element = max(enumerate(scores_tmp), key=itemgetter(1))
# if scores1[i][2] >= scores2[i][2]:
# scores.append([attr_i, attrs[attr_i],scores1[i][2],scores1[i][3], scores1[i][4], score_names[0]])
# else:
# scores.append([attr_i, attrs[attr_i],scores2[i][2],scores2[i][3], scores2[i][4], score_names[1]])
score_tmp = weights[0] * scores_tmp[0] + weights[
1] * scores_tmp[1] + weights[2] * scores_tmp[2]
scores.append([
attr_i, attrs[attr_i], score_tmp, scores2[i][3],
scores2[i][4], score_names[index]
])
scores = sorted(scores, key=lambda tup: tup[2])
scores.reverse()
scores_examples = []
for attr_score in scores:
# attr_splt = attr_score[1].split()
ind = attributes_list_orig.index(attr_score[1])
if 'coded_values' not in schema[ind]:
continue
arg_max_examples_vals = schema[ind]['coded_values']
arg_max_examples_vals.sort()
scores_examples.append(attr_score + [arg_max_examples_vals])
top = 0
output = []
for score in scores_examples:
if len(score) <= score_len:
continue
if score[
2] > r.topic_to_attr_threshold and top <= r.topic_to_attr_count:
output.append(score)
top += 1
for match in output:
kb_match_entry = {
'concept': match[3],
'datasource': source_name,
'attribute': match[1],
'match_score': match[2],
'example_values': match[6],
'topic_source': match[5],
'data_type': schema[match[0]]['data_type'],
'score_name': match[4]
}
update_kb_json(m.kbs[source_name], kb_match_entry)
# for debugging:
kb_match_entry['example_values'] = kb_match_entry[
'example_values'][:min(
len(kb_match_entry['example_values']), 5)]
pprint.pprint(kb_match_entry)
with open(p.schema_p, 'w') as fp:
json.dump(m.schema_set, fp, sort_keys=True, indent=2)
with open(kb_curr_file, 'w') as fp:
json.dump(m.kbs, fp, sort_keys=True, indent=2)
print('done saving kb_file', source_name)
print('^^^ PROGRESS', len(m.kbs) / len_all_ds)
return