def generate_sentences(data_files):
"""
Generates SentenceRecord objects using the given datafiles. Tokenises reports into sentences and then preprocesses
each sentence. Also saves the report_id and report class into the object.
Note: Currently report class is just the index of the datafile extracted from, a global mapping would be better.
:param data_files: List of strings corresponding the raw csv datafiles in the format 'report_id, report'
:return: List of SentenceRecord objects extracted from the files.
"""
list_sentence_record = []
tokenizer = nltk.data.load('tokenizers/punkt/english.pickle')
for i, file_name in enumerate(data_files):
with open(file_name, 'rb') as file:
file.readline() # skip header line
reader = csv.reader(file)
for report in reader:
for sentence in tokenizer.tokenize(report[1]):
tmp = SentenceRecord(sentence)
tmp.processed_sentence = " ".join(
preprocess.textPreprocess(sentence, removeNegationsFromSentences=False))
tmp.report_id = report[0]
tmp.report_class = i
list_sentence_record.append(tmp)
return list_sentence_record
def search(model, numResults, searchTerm):
dictionary = gensim.corpora.Dictionary.load('./model_files/reports.dict')
origSearchTerm = searchTerm
searchTerm = preprocess.textPreprocess(searchTerm)
# searchTerm = preprocess.getDerivations(searchTerm)
if (searchTerm == []):
return []
if model == "bow":
index = gensim.similarities.SparseMatrixSimilarity.load('./model_files/reports.index')
index.num_best = numResults
searchTerm_bow = dictionary.doc2bow(searchTerm)
similarReports = index[searchTerm_bow]
elif model == "tfidf":
tfidf_model = gensim.models.TfidfModel.load('./model_files/reports.tfidf_model')
tfidf_index = gensim.similarities.SparseMatrixSimilarity.load('./model_files/reports_tfidf.index')
tfidf_index.num_best = numResults
searchTerm_bow = dictionary.doc2bow(searchTerm)
searchTerm_tfidf = tfidf_model[searchTerm_bow]
similarReports = tfidf_index[searchTerm_tfidf]
elif model == "lsi":
tfidf_model = gensim.models.TfidfModel.load('./model_files/reports.tfidf_model')
lsi_model = gensim.models.LsiModel.load('./model_files/reports.lsi_model')
lsi_index = gensim.similarities.MatrixSimilarity.load('./model_files/reports_lsi.index')
lsi_index.num_best = numResults
searchTerm_bow = dictionary.doc2bow(searchTerm)
searchTerm_tfidf = tfidf_model[searchTerm_bow]
searchTerm_lsi = lsi_model[searchTerm_tfidf]
similarReports = lsi_index[searchTerm_lsi]
elif model == "lda":
lda_model = gensim.models.LdaModel.load('./model_files/reports.lda_model')
lda_index = gensim.similarities.MatrixSimilarity.load('./model_files/reports_lda.index')
lda_index.num_best = numResults
searchTerm_bow = dictionary.doc2bow(searchTerm)
searchTerm_lda = lda_model[searchTerm_bow]
similarReports = lda_index[searchTerm_lda]
elif model == "doc2vec":
model = gensim.models.Doc2Vec.load("./model_files/reports.doc2vec_model")
searchTerm_docvec = model.infer_vector(searchTerm)
similarReports = model.docvecs.most_similar([searchTerm_docvec],topn=numResults)
elif model == "rnn":
searchTerm_rnn = rnn.getReportSearchTerm(origSearchTerm)
similarReports = rnn.most_similar_reports(searchTerm_rnn,topn=numResults)
else:
return 0
return similarReports
def input_game():
"""
This game accepts a 'report sentence' from a user and attempts to classify it.
:return: None
"""
print "Write a sentence to see if the system thinks it's diagnostic:"
sys.stdout.write("> ")
ans = sys.stdin.readline()
print ans
if ans and ans.rstrip() != "":
ans = ans.rstrip()
processed_sentence = " ".join(
preprocess.textPreprocess(ans, removeNegationsFromSentences=False))
if processed_sentence == "":
return
print processed_sentence
probs = pipe.predict_proba([processed_sentence])[0]
print probs
if probs[0] > probs[1]:
print "[Negative] That output is not diagnostic with a confidence of " + str(probs[0])
else:
print "[Positive] That output is diagnostic with a confidence of " + str(probs[1])
