def predict_case(docket_number, leaning=None):
log = logging.getLogger('PREDICT_CASE')
if not leaning:
leaning = scdb.get_case_info(docket_number)["decisionDirection"]
the_transcript = transcript.get_transcript_from_PDF("transcripts/" + docket_number + ".pdf")
petitioners, respondents = transcript.get_petitioners_and_respondents(the_transcript)
argument = transcript.get_argument(the_transcript)
arguments_by_advocate = transcript.get_arguments_by_advocate(petitioners, respondents, argument)
inputs = []
for petitioner, argument in arguments_by_advocate["petitioner"].iteritems():
statements = transcript.get_statements_in_argument(argument, petitioner)
number_of_words_per_speaker = scores.get_number_of_words_per_speaker(statements)
features = scores.get_features_from_statements(statements)
flat_features = scores.flatten_features(features)
normalized = scores.normalize_feature_list(flat_features)
myscores = scores.get_feature_vector(normalized)
inputs.extend(myscores)
for respondent, argument in arguments_by_advocate["respondent"].iteritems():
statements = transcript.get_statements_in_argument(argument, respondent)
number_of_words_per_speaker = scores.get_number_of_words_per_speaker(statements)
features = scores.get_features_from_statements(statements)
flat_features = scores.flatten_features(features)
normalized = scores.normalize_feature_list(flat_features)
myscores = scores.get_feature_vector(normalized)
inputs.extend(myscores)
if len(inputs) < 774:
# We were unable to match all arguments
# Not gonna help us, continue
print "Parsing error? Did not create enough features."
return 0
if len(inputs) > 774:
# More than one argument per advocate
# just get the first two
inputs = inputs[:774]
print "Parsing error? Had to truncate features."
inputs.append(leaning)
log.info(inputs[:25])
log.info("Querying network. . .")
return net.activate(inputs)[0]
def inline_predict(docket_number):
info = scdb.get_case_info(docket_number)
result = predict_case(docket_number)
resultint = int(round(result))
if resultint > 1:
resultint = 1
if resultint == 1:
winner = "Petitioner"
else:
winner = "Respondent"
if info["partyWinning"] == "1":
actual = "Petitioner"
elif info["partyWinning"] == "0":
actual = "Respondent"
else:
actual = "Unclear"
correct = False
if winner == actual:
correct = True
confidence_level = int(abs(result - 0.5)*100)
if confidence_level < 40:
low_confidence = True
return render_template('inline_predict.html', **locals())
net = buildNetwork(775, 9, 1)
for file in files:
print "Now processing " + file
slash = file.find('/')
end_docket = file.find('_')
if end_docket == -1:
end_docket = file.find(".pdf")
questionnumber = re.search(r"q\d", file)
if questionnumber:
end_docket = questionnumber.start()
docket_number = file[slash+1:end_docket]
print "Checking database for docket #%s" % docket_number
info = scdb.get_case_info(docket_number)
if not info:
print "Couldn't find case in SCDB. Will skip."
continue
print "Now processing transcript for ", info["caseName"]
winner = scdb.get_winning_party(docket_number)
if winner == "unclear":
print "The winner of this case is unclear. Will skip."
continue
print "Winner identified: %s" % winner
print "Decision direction: %s" % info["decisionDirection"]
the_transcript = transcript.get_transcript_from_PDF(file)
petitioners, respondents = transcript.get_petitioners_and_respondents(the_transcript)
argument = transcript.get_argument(the_transcript)
def predict(docket_number):
info = scdb.get_case_info(docket_number)
return render_template('predict.html', **locals())
def get_case_info(docket_number):
return jsonify(scdb.get_case_info(docket_number))