import json import pandas as pd # load manual check sheet df_predicate = pd.read_csv('predicate_manual_check.csv') df_predicate = df_predicate[df_predicate['remove (noisy)'] != 'x'] df_predicate = df_predicate[df_predicate['remove (too vague)'] != 'x'] predicate_main = df_predicate[df_predicate['ok'] == 'x']['unique predicates'].tolist() df_sub = df_predicate[df_predicate['ok'] != 'x'] df_sub_same = df_sub[['unique predicates', 'same as']].dropna() df_sub_same = df_sub_same[[i in predicate_main for i in df_sub_same['same as']]] df_sub_same.index = df_sub_same.pop('unique predicates') sub_same = df_sub_same['same as'].to_dict() df_sub_rev = df_sub[['unique predicates', 'reverse of']].dropna() df_sub_rev = df_sub_rev[[i in predicate_main for i in df_sub_rev['reverse of']]] df_sub_rev.index = df_sub_rev.pop('unique predicates') sub_rev = df_sub_rev['reverse of'].to_dict() # load data and filter based on manual predicate check sheet with open(f"data/t_rex.filter.jsonl") as f: data = pd.DataFrame([json.loads(i) for i in f.read().split('\n') if len(i) > 0]) data['predicate'] = [sub_same[i] if i in sub_same else i for i in data['predicate']] data['reverse'] = [i in sub_rev for i in data['predicate']] data['predicate'] = [sub_rev[i] if i in sub_rev else i for i in data['predicate']] data_remain = data[[i not in predicate_main for i in data['predicate']]] data_remain = data_remain[[i.startswith("P") for i in data_remain['predicate']]] data = data[[i in predicate_main for i in data['predicate']]] data_rev = data[data['reverse']].copy() o = data_rev.pop("object") s = data_rev.pop("subject") data_rev["subject"] = o data_rev["object"] = s data[data['reverse']] = data_rev[data.columns] data.pop("reverse") df_main = df_predicate[df_predicate['ok'] == 'x'][['unique predicates', 'pretty relation name', 'pretty relation name is reverse']] df_main['reverse'] = [i == 'x' for i in df_main.pop('pretty relation name is reverse')] df_main['predicate'] = df_main.pop('unique predicates') data_filter_join = data.merge(df_main, how='inner', on='predicate') data_filter_join_rev = data_filter_join[data_filter_join['reverse']].copy() o = data_filter_join_rev.pop("object") s = data_filter_join_rev.pop("subject") data_filter_join_rev["subject"] = o data_filter_join_rev["object"] = s data_filter_join[data_filter_join['reverse']] = data_filter_join_rev[data_filter_join.columns] data_filter_join.pop("reverse") data_filter_join.pop("predicate") data_filter_join['predicate'] = data_filter_join.pop("pretty relation name") new = pd.concat([data_filter_join, data_remain[data_filter_join.columns]]) print(f"{len(new)} triples, {len(new['predicate'].unique())} predicates") with open(f"data/t_rex.filter_unified.jsonl", 'w') as f: for _, i in new.iterrows(): f.write(json.dumps(i.to_dict()) + '\n')