class NLC(object):
def __init__(self, credential_file_path=None):
self.__nlc = None
self.__initialize(credential_file_path)
def __initialize(self, credential_file_path):
if not credential_file_path:
credential_file_path = os.path.expanduser(DEFAULT_CREDENTIAL_PATH)
with open(credential_file_path, 'r') as credential_file:
credential = json.load(credential_file)
self.__nlc = NaturalLanguageClassifier(url=credential['url'], username=credential['username'], password=credential['password'])
def create(self, traning_data, name=None, language='en'):
"""
:param traning_data: A csv file or file path representing the traning data
:param name: The optional descriptive name for the classifier
:param language: The language og the input data
:return: A instance object with the classifier_id of the newly created classifier, still in traning
"""
create_result = None
if isinstance(traning_data, file) or isinstance(traning_data, IOBase): # traning_data is file discripter
create_result = self.__nlc.create(traning_data, name=name, language=language)
elif isinstance(traning_data, str): # traning_data is file path
with open(traning_data, newline=None, mode='r', encoding='utf-8') as csv_file:
if is_valid_recode_num(csv_file):
create_result = self.__nlc.create(csv_file, name=name, language=language)
return CreateResult(create_result)
def classifiers(self):
classifiers_raw = self.__nlc.list()
classifiers_ = [Classifier(c) for c in classifiers_raw['classifiers']]
return Classifiers(classifiers_)
def status(self, classifier_id):
return Status(self.__nlc.status(classifier_id))
def classify(self, classifier_id, text):
return ClassifyResult(self.__nlc.classify(classifier_id, text))
def remove(self, classifier_id):
"""
param: classifier_id: Unique identifier for the classifier
retrun: empty dict object
raise: watson_developer_cloud.watson_developer_cloud_service.WatsonException: Not found
"""
return self.__nlc.remove(classifier_id)
def remove_all(self):
classifiers_ = self.classifiers()
return [self.remove(c.classifier_id) for c in classifiers_]
def train_nlc(url, username, password, truth, name):
logger.info("Train model %s with %d instances" % (name, len(truth)))
with tempfile.TemporaryFile() as training_file:
# NLC cannot handle newlines.
truth[QUESTION] = truth[QUESTION].str.replace("\n", " ")
to_csv(training_file, truth[[QUESTION, ANSWER_ID]], header=False, index=False)
training_file.seek(0)
nlc = NaturalLanguageClassifier(url=url, username=username, password=password)
r = nlc.create(training_data=training_file, name=name)
logger.info(pretty_print_json(r))
return r["classifier_id"]
def train_nlc(url, username, password, truth, name):
logger.info("Train model %s with %d instances" % (name, len(truth)))
with tempfile.TemporaryFile() as training_file:
# NLC cannot handle newlines.
truth[QUESTION] = truth[QUESTION].str.replace("\n", " ")
to_csv(training_file, truth[[QUESTION, ANSWER_ID]], header=False, index=False)
training_file.seek(0)
nlc = NaturalLanguageClassifier(url=url, username=username, password=password)
r = nlc.create(training_data=training_file, name=name)
logger.info(pretty_print_json(r))
return r["classifier_id"]
class Watson_api():
def __init__(self):
self.fname = ""
self.modelSearchList = ModelSearchList()
self.text_data = []
self.target_label = []
self.watson_crediantial = watson_key()
#self.watson_classifier = self.watson_crediantial.classifier_twitter_classfier
#self.watson_classifier = self.watson_crediantial.classifier_twitter_hash_classfier
#self.watson_classifier = self.watson_crediantial.classifier_twitter_unblance_keyword_classfier
#self.watson_classifier = self.watson_crediantial.twitter_priority_classfier
self.watson_classifier = self.watson_crediantial.twitter_category_classfier
self.natural_language_classifier = NaturalLanguageClassifier(username=self.watson_crediantial.username,
password=self.watson_crediantial.password)
#print(json.dumps(self.natural_language_classifier.list(), indent=2))
def parse_args(self):
p = ArgumentParser(description='Encoder-decoder neural machine trainslation')
p.add_argument('data', help='[in] data')
args = p.parse_args()
return args
def train(self):
# create a classifier
with open('../resources/weather_data_train.csv', 'rb') as training_data:
print(json.dumps(self.natural_language_classifier.create(training_data=training_data, name='weather2'), indent=2))
def __read_data(self):
for line in open(self.fname, "r"):
split_line = line.split(",")
self.text_data.append(split_line[0].strip())
self.target_label.append(self.modelSearchList.search_category_dictionary[split_line[1].strip()])
def predict(self, args):
# replace 47C164-nlc-243 with your classifier id
status = self.natural_language_classifier.status(self.watson_classifier)
self.fname = args.data
self.__read_data()
predict_id = []
#print (json.dumps(status, indent=2, ensure_ascii=False))
for i in range(len(self.text_data)):
classes = self.natural_language_classifier.classify(self.watson_classifier, self.text_data[i])
class_id = self.modelSearchList.search_category_dictionary[classes["classes"][0]["class_name"].replace("\"", "").replace("\"", "")]
predict_id.append(class_id)
print(self.target_label)
print(predict_id)
f1_score_twitter = f1_score(self.target_label, predict_id, average='macro')
print("----F measure-----")
print(f1_score_twitter)
def nlc_router_train(url, username, password, oracle_out, path, all_correct):
"""
NLC Training on the oracle experiment output to determine which system(NLC or Solr) should
answer particular question.
1. Splitting up the oracle experiment output data into 8 equal training records and testing records. This is to
ensure 8-fold cross validation of the data-set. All training and Testing files will be stored
at the "path"
2. Perform NLC training on the all 8 training set simultaneously and returns list of classifier
ids as json file in the working directory
:param url: URL of NLC instance
:param username: NLC Username
:param password: NLC password
:param oracle_out: file created by oracle experiment
:param path: directory path to save intermediate results
:param all_correct: optional boolean parameter to train with only correct QA pairs
:return: list of classifier ids by NLC training
"""
ensure_directory_exists(path)
sys_name = oracle_out[SYSTEM][0]
oracle_out[QUESTION] = oracle_out[QUESTION].str.replace("\n", " ")
kfold_split(oracle_out, path, NLC_ROUTER_FOLDS, True)
classifier_list = []
list = []
for x in range(0, NLC_ROUTER_FOLDS):
train = pandas.read_csv(os.path.join(path, "Train{0}.csv".format(str(x))))
if all_correct:
logger.info("Training only on CORRECT examples.")
# Ignore records from training which are not correct
train = train[train[CORRECT]]
train = train[train[IN_PURVIEW]]
train = train[[QUESTION, ANSWERING_SYSTEM]]
logger.info("Training set size = {0}".format(str(len(train))))
with tempfile.TemporaryFile() as training_file:
to_csv(training_file, train[[QUESTION, ANSWERING_SYSTEM]], header=False, index=False)
training_file.seek(0)
nlc = NaturalLanguageClassifier(url=url, username=username, password=password)
classifier_id = nlc.create(training_data=training_file, name="{0}_fold_{1}".format(str(sys_name), str(x)))
classifier_list.append(classifier_id["classifier_id"].encode("utf-8"))
list.append({classifier_id["name"].encode("utf-8"): classifier_id["classifier_id"].encode("utf-8")})
logger.info(pretty_print_json(classifier_id))
pretty_print_json(classifier_id)
with open(os.path.join(path, 'classifier.json'), 'wb') as f:
json.dump(list, f)
return classifier_list
class NLClassifier(object):
def __init__(self, username, password, classifier):
# Setup Watson SDK
self.natural_language_classifier = NLC(username=username,password=password)
# Classifier information
self.classifier = {}
self.classifier['name'] = classifier['name']
self.classifier['training_file'] = classifier['training_file']
c = self.natural_language_classifier.list_classifiers()
if any(d['name'] == self.classifier['name'] for d in c['classifiers'] ):
self.classifier['id'] = [ d['classifier_id'] for d in c['classifiers'] if d['name'] == self.classifier['name'] ][0]
print 'Found classifier id %s ' % self.classifier['id']
self.classifier['status'] = self.natural_language_classifier.status(self.classifier['id'])['status']
else:
print 'No classifier found, creating new from training set'
self.classifier['id'] = self.create_classifier()
print 'New classifier id: %s ' % self.classifier['id']
### Method to train the Watson Natural Language Classifier
# The training set is delivered as a CSV file as specified in the Developer Guide
# https://www.ibm.com/watson/developercloud/doc/nl-classifier/data_format.shtml
def create_classifier(self):
training_data = open(self.classifier['training_file'], 'rb')
training_result = self.natural_language_classifier.create( training_data=training_data, name=self.classifier['name'] )
if training_result['status'] == "Training":
self.classifier['status'] = "Training"
return training_result['classifier_id']
else:
print training_result
return "Error"
def classify(self,text):
# Typically in a production system Watson NLC will be fully trained and verified by a data scientist before the system is ever
# exposed in production. However because this is a demo application where Watson NLC is trained at application deployment time,
# we will need to have a check to verify that the training is completed.
if self.classifier['status'] == "Training":
r = self.natural_language_classifier.status(self.classifier['id'])
if r['status'] == "Training":
return {"error": "Classifier still in training. Please try again in a few minutes."}
elif r['status'] == "Available":
self.classifier['status'] = 'Available'
else:
return {"error": "Unknown status for classifier", "message": r['status']}
return self.natural_language_classifier.classify(self.classifier['id'], text)
#! /usr/bin/python
from watson_developer_cloud import NaturalLanguageClassifierV1 as NLC
import json
with open('credential.json') as f_cred:
cred = json.load(f_cred)
nlc = NLC(username = cred['username'], password = cred['password'])
with open('weather_data_train.csv') as f_train:
clsfier = nlc.create(
training_data = f_train,
name = 'python classfier',
language = 'en')
with open ('classifier_info.json', 'w') as f_cls:
json.dump(clsfier, f_cls, indent = 2)
print(json.dumps(clsfier, indent = 2))
class Watson_api():
def __init__(self):
self.fname = ""
self.modelSearchList = ModelSearchList()
self.text_data = []
self.target_label = []
self.watson_crediantial = watson_key()
#self.watson_classifier = self.watson_crediantial.classifier_twitter_classfier
#self.watson_classifier = self.watson_crediantial.classifier_twitter_hash_classfier
#self.watson_classifier = self.watson_crediantial.classifier_twitter_unblance_keyword_classfier
#self.watson_classifier = self.watson_crediantial.twitter_priority_classfier
self.watson_classifier = self.watson_crediantial.twitter_category_classfier
self.natural_language_classifier = NaturalLanguageClassifier(
username=self.watson_crediantial.username,
password=self.watson_crediantial.password)
#print(json.dumps(self.natural_language_classifier.list(), indent=2))
def parse_args(self):
p = ArgumentParser(
description='Encoder-decoder neural machine trainslation')
p.add_argument('data', help='[in] data')
args = p.parse_args()
return args
def train(self):
# create a classifier
with open('../resources/weather_data_train.csv',
'rb') as training_data:
print(
json.dumps(self.natural_language_classifier.create(
training_data=training_data, name='weather2'),
indent=2))
def __read_data(self):
for line in open(self.fname, "r"):
split_line = line.split(",")
self.text_data.append(split_line[0].strip())
self.target_label.append(
self.modelSearchList.search_category_dictionary[
split_line[1].strip()])
def predict(self, args):
# replace 47C164-nlc-243 with your classifier id
status = self.natural_language_classifier.status(
self.watson_classifier)
self.fname = args.data
self.__read_data()
predict_id = []
#print (json.dumps(status, indent=2, ensure_ascii=False))
for i in range(len(self.text_data)):
classes = self.natural_language_classifier.classify(
self.watson_classifier, self.text_data[i])
class_id = self.modelSearchList.search_category_dictionary[
classes["classes"][0]["class_name"].replace("\"", "").replace(
"\"", "")]
predict_id.append(class_id)
print(self.target_label)
print(predict_id)
f1_score_twitter = f1_score(self.target_label,
predict_id,
average='macro')
print("----F measure-----")
print(f1_score_twitter)
import sys
import operator
import requests
import json
import twitter
from watson_developer_cloud import NaturalLanguageClassifierV1 as NaturalLanguageClassifier
#The IBM Bluemix credentials
nlc_username = '******'
nlc_password = '******'
natural_language_classifier = NaturalLanguageClassifier(
username=nlc_username, password=nlc_password)
with open('../resources/weather_data_train.csv', 'rb') as training_data:
classifier = natural_language_classifier.create(
training_data=training_data,
name='MySampleClassifierPython',
language='en'
)
#print
print(json.dumps(classifier, indent=2))
sys.exit(2)
for opt, arg in opts:
if opt == '-h':
usage()
sys.exit()
elif opt in ("-t", "---trainingdata"):
trainingdata_filepath = arg
elif opt in ("-n", "---name"):
name = arg
elif opt in ("-l", "---language"):
language = arg
elif opt == '-d':
DEBUG = True
if not trainingdata_filepath or not name or not language:
print('Required argument missing.')
usage()
sys.exit(2)
try:
# create classifiers with the training data
natural_language_classifier = NaturalLanguageClassifier(url=nlcConstants.getUrl(), username=nlcConstants.getUsername(), password=nlcConstants.getPassword())
with open(trainingdata_filepath, 'rb') as training_data:
res = natural_language_classifier.create(training_data, name, language)
sys.stdout.write('Response: \n%s\n' % json.dumps(res, indent=2))
except Exception as e:
sys.stdout.write(str(e))
exit(1)
if opt == '-h':
usage()
sys.exit()
elif opt in ("-t", "---trainingdata"):
trainingdata_filepath = arg
elif opt in ("-n", "---name"):
name = arg
elif opt in ("-l", "---language"):
language = arg
elif opt == '-d':
DEBUG = True
if not trainingdata_filepath or not name or not language:
print('Required argument missing.')
usage()
sys.exit(2)
try:
# create classifiers with the training data
natural_language_classifier = NaturalLanguageClassifier(
url=nlcConstants.getUrl(),
username=nlcConstants.getUsername(),
password=nlcConstants.getPassword())
with open(trainingdata_filepath, 'rb') as training_data:
res = natural_language_classifier.create(training_data, name, language)
sys.stdout.write('Response: \n%s\n' % json.dumps(res, indent=2))
except Exception as e:
sys.stdout.write(str(e))
exit(1)
