def gen_ranking_rocs(data_dir, output_dir, query_id, num_topics, keywords): # the directory which contains the training set # (TRUE negatives and TRUE positives) truth_dir = "%s\\%d" % (data_dir, query_id) # Using unormalized keywords project_name = "Q%d-UNW-%dT" % (query_id, num_topics) config_file = "%s\\%s.cfg" % (output_dir, project_name) norm_tokens = keywords try: eval_keywordlda_topiclda_lucene_ranking(project_name, config_file, truth_dir, norm_tokens, output_dir=output_dir) except: print 'Execption in processing', project_name # Using lemmatized keywords project_name = "Q%d-LW-%dT" % (query_id, num_topics) config_file = "%s\\%s.cfg" % (output_dir, project_name) norm_tokens = ' '.join( lemmatize_tokens( regex_tokenizer(keywords) ) ) try: eval_keywordlda_topiclda_lucene_ranking(project_name, config_file, truth_dir, norm_tokens, output_dir=output_dir) except: print 'Execption in processing', project_name # Using Stemming and Lemmatization project_name = "Q%d-LSW-%dT" % (query_id, num_topics) config_file = "%s\\%s.cfg" % (output_dir, project_name) norm_tokens = ' '.join( stem_tokens( lemmatize_tokens( regex_tokenizer(keywords) ) ) ) # try: eval_keywordlda_topiclda_lucene_ranking(project_name, config_file, truth_dir, norm_tokens, output_dir=output_dir) except: print 'Execption in processing', project_name
query_id = 201 file_prefix = '%d-UNT-ALL' % query_id config_file = "F:\\Research\\datasets\\trec2010\\Q201-UNT-30T.cfg" # configuration file, created using the SMARTeR GUI test_directory = "F:\\Research\\datasets\\trec2010\\201"# the directory where we keep the training set (TRUE negatives and TRUE positives) keywords = 'pre-pay swap' norm_tokens = keywords # Unnormalized keywords # eval_keywordlda_topiclda_lucene(file_prefix, config_file, test_directory, norm_tokens) eval_ranking_methods(file_prefix, config_file, test_directory, norm_tokens) query_id = 201 file_prefix = '%d-LW-ALL' % query_id config_file = "F:\\Research\\datasets\\trec2010\\Q201-LW-30T.cfg" # configuration file, created using the SMARTeR GUI test_directory = "F:\\Research\\datasets\\trec2010\\201" # the directory where we keep the training set (TRUE negatives and TRUE positives) keywords = 'pre-pay swap' norm_tokens = ' '.join( lemmatize_tokens( regex_tokenizer(keywords) ) ) # Lemmatization # eval_keywordlda_topiclda_lucene(file_prefix, config_file, test_directory, norm_tokens) eval_ranking_methods(file_prefix, config_file, test_directory, norm_tokens) # query_id = 201 # file_prefix = '%d-LST' % query_id # config_file = "F:\\Research\\datasets\\trec2010\\Q201-LST-30T.cfg" # configuration file, created using the SMARTeR GUI # test_directory = "F:\\Research\\datasets\\trec2010\\201"# the directory where we keep the training set (TRUE negatives and TRUE positives) # keywords = 'pre-pay swap' # norm_tokens = ' '.join( stem_tokens( lemmatize_tokens( regex_tokenizer(keywords) ) ) ) # Stemming and Lemmatization # eval_keywordlda_topiclda_lucene(file_prefix, config_file, test_directory, norm_tokens) # query_id = 207 # file_prefix = '%d-LT' % query_id # config_file = "F:\\Research\\datasets\\trec2010\\Q207-LT-5T.cfg" # configuration file, created using the SMARTeR GUI # test_directory = "F:\\Research\\datasets\\trec2010\\207" # the directory where we keep the training set (TRUE negatives and TRUE positives)
def eval_ranking_varying_topics(query_id, data_dir, keywords, limit = 1000, img_extension = '.eps'): tokens = ' '.join( lemmatize_tokens( regex_tokenizer(keywords) ) ) # Lemmatization lucene_query = 'all:(%s)' % tokens # search in all fields print 'Lucene query:', lucene_query print 'TM query:', tokens truth_dir = "%s%d" % (data_dir, query_id) positive_dir = os.path.join(truth_dir, RELEVANT_DIR_NAME) # TRUE positive documents topiclda_rocs_file_name = '%d-LW-Topic-LDA-VaryingTopics-ROCs' % query_id + img_extension topiclda_rocs_img_title = 'Q%d (Topic-LDA): Varying # of LDA Topics and Lemmas' % query_id keywordlda_rocs_file_name = '%d-LW-Keyword-LDA-VaryingTopics-ROCs' % query_id + img_extension keywordlda_rocs_img_title = 'Q%d (Keyword-LDA): Varying # of LDA Topics and Lemmas' % query_id topics = [5, 10, 15, 20, 30, 40, 50, 60, 70] roc_labels = [] roc_topiclda_list = [] roc_keywordlda_list = [] for idx, num_topics in enumerate(topics): print '------------------------------------------------------------------------------------------' #---------------------------------------------- Reads the configuration file config_file = "%sQ%d-LW-%dT.cfg" % (data_dir, query_id, num_topics) # configuration file, created using the SMARTeR GUI mdl_cfg = read_config(config_file) # Loads the LDA model (lda_dictionary, lda_mdl, lda_index, lda_file_path_index, lda_theta, lda_beta) = load_lda_parameters(mdl_cfg) #------------ Checks whether the keywords are there in the corpus dictionary valid_tokens = 0 for token in tokens.split(): if token.strip() not in lda_dictionary.values(): print token, "is not in the corpus vocabulary." else: valid_tokens += 1 if valid_tokens == 0: print 'None of the tokens exist in the dictionary. Exiting topic search!' exit() # Gets the query topic distribution from the LDA beta print 'Estimated topic dist. from the LDA beta:' query_td2 = get_lda_query_td2(tokens, lda_dictionary, lda_beta) dominant_topics_idx2 = get_query_top_topic_idx(query_td2, lda_mdl, TOP_K_TOPICS) # Gets the query topic distribution from the LDA model print 'Estimated topic dist. from the LDA model:' query_td = get_lda_query_td(tokens, lda_dictionary, lda_mdl) dominant_topics_idx = get_query_top_topic_idx(query_td, lda_mdl, TOP_K_TOPICS) #------------------------------------------------------------- Lucene search if idx == 0: # the first Lucene ranking is added as a reference print 'Lucene ranking' # lu_docs = search_li(lucene_query, limit, mdl_cfg) lu_docs = search_whoosh_index(lucene_query, mdl_cfg) _, lu_docs_list = lu_append_nonresp(lu_docs, truth_dir) lu_res = convert_to_roc_format(lu_docs_list, positive_dir) roc_topiclda_list.append(ROCData(lu_res)) roc_keywordlda_list.append(ROCData(lu_res)) roc_labels.append('Lucene') #---------------------------------------------------------------- LDA search # # Gets the dominant topics from the LDA model # dominant_topics = get_dominant_query_topics(tokens, lda_dictionary, lda_mdl, TOP_K_TOPICS) # dominant_topics_idx = [idx for (idx, _) in dominant_topics] # get the topic indices print 'LDA (w/ keywords) ranking' lda_docs = search_tm2(query_td, lda_index, lda_file_path_index, limit) lda_res = convert_to_roc_format(lda_docs, positive_dir) print 'LDA (w/ keywords) method-2 ranking' lda_docs2 = search_tm2(query_td2, lda_index, lda_file_path_index, limit) lda_res2 = convert_to_roc_format(lda_docs2, positive_dir) print 'LDA (w/ query topics) ranking' lda_tts_docs = search_tm_topics(dominant_topics_idx, limit, lda_file_path_index, lda_theta) lda_tts_res = convert_to_roc_format(lda_tts_docs, positive_dir) print 'LDA (w/ query topics) method-2 ranking' lda_tts_docs2 = search_tm_topics(dominant_topics_idx2, limit, lda_file_path_index, lda_theta) lda_tts_res2 = convert_to_roc_format(lda_tts_docs2, positive_dir) roc_topiclda_list.append(ROCData(lda_tts_res)) roc_keywordlda_list.append(ROCData(lda_res)) roc_labels.append('%d topics' % num_topics) roc_topiclda_list.append(ROCData(lda_tts_res2)) roc_keywordlda_list.append(ROCData(lda_res2)) roc_labels.append('%d topics (method-2)' % num_topics) print '------------------------------------------------------------------------------------------' ## Plot ROC curves plot_multiple_roc(roc_topiclda_list, title=topiclda_rocs_img_title, labels=roc_labels, include_baseline=True, file_name=topiclda_rocs_file_name) plot_multiple_roc(roc_keywordlda_list, title=keywordlda_rocs_img_title, labels=roc_labels, include_baseline=True, file_name=keywordlda_rocs_file_name)
def eval_ranking_varying_topics(query_id, dir_path, keywords, limit = 1000, img_extension = '.eps'): tm_query = ' '.join( lemmatize_tokens( regex_tokenizer(keywords) ) ) # Lemmatization lucene_query = 'all:(%s)' % tm_query # search in all fields print 'Lucene query:', lucene_query print 'TM query:', tm_query test_directory = "%s%d" % (dir_path, query_id) positive_dir = os.path.join(test_directory, "1") # TRUE positive documents TOP_K_TOPICS = 5 # the number topics used for Topic-LDA topiclda_rocs_file_name = '%d-LT-Topic-LDA-VaryingTopics-ROCs' % query_id + img_extension topiclda_rocs_img_title = 'Q%d Topic-LDA with Varying Number of Topics' % query_id keywordlda_rocs_file_name = '%d-LT-Keyword-LDA-VaryingTopics-ROCs' % query_id + img_extension keywordlda_rocs_img_title = 'Q%d Keyword-LDA with Varying Number of Topics' % query_id topics = [5, 10, 15, 20, 30, 40, 50, 60, 70, 80] roc_labels = [] roc_topiclda_list = [] roc_keywordlda_list = [] for idx, num_topics in enumerate(topics): #---------------------------------------------- Reads the configuration file config_file = "%sQ%d-LT-%dT.cfg" % (dir_path, query_id, num_topics) # configuration file, created using the SMARTeR GUI mdl_cfg = read_config(config_file) # Loads the LDA model lda_dictionary, lda_mdl, _, _ = load_tm(mdl_cfg) #------------ Checks whether the keywords are there in the corpus dictionary valid_tokens = 0 for token in tm_query.split(): if token.strip() not in lda_dictionary.values(): print token, "is not in the corpus vocabulary. Hence, this word will be ignored from the topic search." else: valid_tokens += 1 if valid_tokens == 0: print 'None of the tokens exist in the dictionary. Exiting topic search!' exit() #------------------------------------------------------------- Lucene search if idx == 0: # the first Lucene ranking is added as a reference print 'Lucene ranking' lu_docs = search_li(lucene_query, limit, mdl_cfg) _, lu_docs_list = lu_append_nonresp(lu_docs, test_directory) lu_res = convert_to_roc_format(lu_docs_list, positive_dir) roc_topiclda_list.append(ROCData(lu_res)) roc_keywordlda_list.append(ROCData(lu_res)) roc_labels.append('Lucene') #---------------------------------------------------------------- LDA search # Gets the dominant topics from the LDA model dominant_topics = get_dominant_query_topics(tm_query, lda_dictionary, lda_mdl, TOP_K_TOPICS) dominant_topics_idx = [idx for (idx, _) in dominant_topics] # get the topic indices print 'LDA (w/ keywords) ranking' lda_docs = search_tm(tm_query, limit, mdl_cfg) lda_res = convert_to_roc_format(lda_docs, positive_dir) print 'LDA (w/ query topics) ranking' lda_tts_docs = search_tm_topics(dominant_topics_idx, limit, mdl_cfg) lda_tts_res = convert_to_roc_format(lda_tts_docs, positive_dir) roc_topiclda_list.append(ROCData(lda_tts_res)) roc_keywordlda_list.append(ROCData(lda_res)) roc_labels.append('%d topics' % num_topics) ## Plot ROC curves plot_multiple_roc(roc_topiclda_list, title=topiclda_rocs_img_title, labels=roc_labels, include_baseline=True, file_name=topiclda_rocs_file_name) plot_multiple_roc(roc_keywordlda_list, title=keywordlda_rocs_img_title, labels=roc_labels, include_baseline=True, file_name=keywordlda_rocs_file_name)