def get_runs_api(): global runs_df if (runs_df.empty or request.args.get('topic') is None): return make_response( "", 404, ) else: topic = int(request.args.get('topic')) # dataframe filtered by topic runs_filtered = get_runs.by_topic(runs_df, topic) # shorten the run size to avoid sending too large data runs_filtered = get_runs.by_run_size(runs_filtered, 100) # dataframe with index ['RANK', 'RUN'] runs_by_rank = get_runs.indexed_by_rank(runs_filtered) response = make_response( jsonify(convert_data.get_dict_from_df(runs_by_rank)), 200, ) response.headers["Content-Type"] = "application/json" return response
def get_qrels_api(): global qrels_df if (qrels_df.empty or request.args.get('topic') is None): return make_response( "", 404, ) else: topic = int(request.args.get('topic')) # dataframe filtered by topic and indexed by the document qrels_filtered = get_qrels.by_topic(qrels_df, topic) # dataframe with index ['DOCUMENT'] qrels_by_rank = get_qrels.indexed_by_document(qrels_filtered) response = make_response( jsonify(convert_data.get_dict_from_df(qrels_by_rank)), 200, ) response.headers["Content-Type"] = "application/json" return response
def round_robin_alg(runs, qrels, pool_size, log_path): # identify the names of all unique runs, and how many there are runs_ids = runs["RUN"].unique().tolist() runs_ids.sort() runs_count = len(runs_ids) # keep track of which run we're retrieving from next_run_index = -1 ### HELPER STRUCTURES # dict containing all occurrences of all documents docs_occurrences = adjudication.get_occurrences(runs, log_path) # dict keeping track of which docs have been retrieved from where, and if they are unique retrieved_docs = {} # dict keeping track of how many docs have been retrieved from each run runs_status = {} for item in runs_ids: runs_status[item] = 1 # pandas' dataframe keeping track of how many docs for each run are relevant/unique relevant/nonrelevant run_relevancies = pd.DataFrame(columns=['RUN', 'DOCS_RETRIEVED', 'REL', 'REL_UNIQUE', 'NON_REL']) run_relevancies.set_index('RUN', inplace=True) ### OUTPUT STRUCTURES retrieved_docs_order = [] # ordered array of each doc retrieved at each step run_relevancies_order = [] # ordered array of the relevancies status at each step print("➡️ Retrieving documents to be adjudicated...") for i in range(1, pool_size+1): # repeat the process until we have retrieved the desired number of documents # print("\n### DOCUMENT", i) ### LOGGING # logging.debug("\n#### DOCUMENT %d", i) ### LOGGING next_doc_found = False while not next_doc_found: # find next_run next_run, next_run_index = round_robin.get_next_run(runs_ids, runs_status, next_run_index) # print("next_run", next_run) ### LOGGING # logging.debug("next_run:: %s", next_run) ### LOGGING while True: # find next_doc next_doc = adjudication.get_next_doc(runs, next_run, runs_status, retrieved_docs) if next_doc.empty: # if it's empty, next_run must have run out of documents # update runs_status to track that runs_status[next_run] = -1 # print("Run", next_run, "is empty. Trying again.") ### LOGGING # logging.debug("Run %s appears to be empty. Trying again.", next_run) ### LOGGING # break to go look for another run break # since the proposed next_doc is not empty, we can gather its info next_doc_info = { 'DOCUMENT': next_doc.iloc[0].DOCUMENT, 'RANK': next_doc.iloc[0].RANK } ''' LOGGING ''' ''' LOGGING ''' ''' LOGGING ''' # print("next_doc:: DOCUMENT:", next_doc_info['DOCUMENT'], "RANK:", next_doc_info['RANK'] ) # logging.debug("next_doc:: DOCUMENT: %s RANK %s", next_doc_info['DOCUMENT'], next_doc_info['RANK']) ''' LOGGING ''' ''' LOGGING ''' ''' LOGGING ''' if next_doc_info['DOCUMENT'] not in retrieved_docs: # if the document found has not been retrieved before next_doc_found = True break # otherwise doc has already been retrieved # print("This document is already retrieved. Retrieving a new one:") ### LOGGING # logging.debug("This document is already retrieved. Retrieving a new one:") ### LOGGING # update run_relevancies to track which documents are unique relevants adjudication.check_for_unique_relevants(next_doc_info, retrieved_docs, run_relevancies) # loop to look for another next_doc # finally new document is retrieved # get corresponding qrel next_doc_info['RELEVANCY'] = get_doc_info.relevancy(qrels, next_doc_info['DOCUMENT']) # track which documents have already been retrieved retrieved_docs[next_doc_info['DOCUMENT']] = { 'RELEVANCY': next_doc_info['RELEVANCY'], 'RETRIEVED_FROM': next_run, 'UNIQUE': 1 } # update info for first graph retrieved_docs_order.append({ "DOCUMENT": next_doc_info['DOCUMENT'], "RANK": next_doc_info['RANK'], "RELEVANCY": next_doc_info['RELEVANCY'], "RETRIEVED_FROM": next_run, "OCCURRENCES": docs_occurrences[next_doc_info['DOCUMENT']] }) # update info for second graph run_relevancies = adjudication.update_run_relevancies(run_relevancies, next_run, next_doc_info) run_relevancies_order.append(convert_data.get_dict_from_df(run_relevancies)) convert_data.write_dict_into_json(runs_status, log_path+'adjudication/round_robin/', "runs_status.json", 0) convert_data.write_dict_into_json(retrieved_docs, log_path+'adjudication/round_robin/', "retrieved_docs.json", 0) convert_data.write_dict_into_json(retrieved_docs_order, log_path+'adjudication/round_robin/', "retrieved_docs_order.json", 0) convert_data.write_dict_into_json(convert_data.get_dict_from_df(run_relevancies), log_path+'adjudication/round_robin/', "run_relevancies.json", 0) convert_data.write_dict_into_json(run_relevancies_order, log_path+'adjudication/round_robin/', "run_relevancies_order.json", 0) print("✅ Complete.") return { 'retrieved_docs_order': retrieved_docs_order, 'run_relevancies_order': run_relevancies_order }
def round_robin_alg_og(runs, qrels, pool_size, log_path): # identify the names of all unique runs, and how many there are runs_ids = runs["RUN"].unique().tolist() runs_ids.sort() runs_count = len(runs_ids) print(runs_ids) ### HELPER STRUCTURES # dict keeping track of which docs have been retrieved from where, and if they are unique retrieved_docs = {} # dict keeping track of how many docs have been retrieved from each run runs_status = {} for item in runs_ids: runs_status[item] = 1 # pandas' dataframe keeping track of how many docs for each run are relevant/unique relevant/nonrelevant run_relevancies = pd.DataFrame( columns=['RUN', 'DOCS_RETRIEVED', 'REL', 'REL_UNIQUE', 'NON_REL']) run_relevancies.set_index('RUN', inplace=True) ### OUTPUT STRUCTURES retrieved_docs_order = [ ] # ordered array of each doc retrieved at each step run_relevancies_order = [ ] # ordered array of the relevancies status at each step print("➡️ Retrieving documents to be adjudicated...") for i in range(1, pool_size + 1): # repeat the process until we have retrieved the desired number of documents # print("Retrieving document", i, "\n") next_run = round_robin.get_next_run_og(runs_ids) next_doc = round_robin.get_next_doc(runs, next_run, runs_status, retrieved_docs) # proposed next_doc next_doc_info = { 'DOCUMENT': next_doc.iloc[0].DOCUMENT, 'RANK': next_doc.iloc[0].RANK } logging.debug("#### DOCUMENT %d", i) logging.debug("next_run: %s", next_run) logging.debug("DOCUMENT: %s \n RANK %s", next_doc_info['DOCUMENT'], next_doc_info['RANK']) while next_doc_info['DOCUMENT'] in retrieved_docs: # if document was already retrieved # update run_relevancies to track which documents are unique relevants round_robin.check_for_unique_relevants(next_doc_info, retrieved_docs, run_relevancies) # find new proposed next_doc next_doc = round_robin.get_next_doc(runs, next_run, runs_status, retrieved_docs) logging.debug( "This document is already retrieved. Retrieving a new one:") logging.debug("DOCUMENT: %s \n RANK %s", next_doc_info['DOCUMENT'], next_doc_info['RANK']) while (next_doc.empty): print("run", next_run, "is empty. Trying again.") logging.debug("Run %s appears to be empty. Trying again.", next_run) # current next_run must have run out of documents runs_ids.pop(len(runs_ids) - 1) next_run = round_robin.get_next_run_og(runs_ids) print("next_run", next_run) logging.debug("next_run: %s", next_run) next_doc = round_robin.get_next_doc(runs, next_run, runs_status, retrieved_docs) logging.debug("DOCUMENT: %s \n RANK %s", next_doc_info['DOCUMENT'], next_doc_info['RANK']) next_doc_info = { 'DOCUMENT': next_doc.iloc[0].DOCUMENT, 'RANK': next_doc.iloc[0].RANK } # finally new document is retrieved # get corresponding qrel next_doc_info['RELEVANCY'] = get_doc_info.relevancy( qrels, next_doc_info['DOCUMENT']) # track which documents have already been retrieved retrieved_docs[next_doc_info['DOCUMENT']] = { 'RELEVANCY': next_doc_info['RELEVANCY'], 'RETRIEVED_FROM': next_run, 'UNIQUE': 1 } # update info for first graph retrieved_docs_order.append({ "DOCUMENT": next_doc_info['DOCUMENT'], "RANK": next_doc_info['RANK'], "RELEVANCY": next_doc_info['RELEVANCY'], "RETRIEVED_FROM": next_run, "OCCURRENCES": get_doc_info.occurrences(runs, next_doc_info['DOCUMENT']) }) # update info for second graph run_relevancies = round_robin.update_run_relevancies( run_relevancies, next_run, next_doc_info) run_relevancies_order.append( convert_data.get_dict_from_df(run_relevancies)) convert_data.write_dict_into_json( runs_status, log_path + 'adjudication/round_robin_og/', "runs_status.json", 0) convert_data.write_dict_into_json( retrieved_docs, log_path + 'adjudication/round_robin_og/', "retrieved_docs.json", 0) convert_data.write_dict_into_json( retrieved_docs_order, log_path + 'adjudication/round_robin_og/', "retrieved_docs_order.json", 0) convert_data.write_dict_into_json( convert_data.get_dict_from_df(run_relevancies), log_path + 'adjudication/round_robin_og/', "run_relevancies.json", 0) convert_data.write_dict_into_json( run_relevancies_order, log_path + 'adjudication/round_robin_og/', "run_relevancies_order.json", 0) print("✅ Complete.") return { 'retrieved_docs_order': retrieved_docs_order, 'run_relevancies_order': run_relevancies_order }
topic = 401 ### RUNS # dataframe filtered by topic runs_filtered = get_runs.by_topic(runs_df, topic) # print(runs_filtered) # shorten the run size to avoid sending too large data runs_filtered = get_runs.by_run_size(runs_filtered, 100) # dataframe with index ['RANK', 'RUN'] runs_by_rank = get_runs.indexed_by_rank(runs_filtered) print(runs_by_rank) runs_by_rank_dict = convert_data.get_dict_from_df(runs_by_rank) convert_data.write_dict_into_json(runs_by_rank_dict, paths['output'] + 'latex/', 'runs_by_rank.json', 1) ### QRELS # dataframe filtered by topic and indexed by the document qrels_filtered = get_qrels.by_topic(qrels_df, topic) # print(qrels_filtered) # dataframe with index ['DOCUMENT'] qrels_by_doc = get_qrels.indexed_by_document(qrels_filtered) # print(qrels_by_doc) qrels_by_doc_dict = convert_data.get_dict_from_df(qrels_by_doc) convert_data.write_dict_into_json(qrels_by_doc_dict,