def setUp(self):
# create a dummy ClassifierModel object
self.classifier_model = ClassifierModel.objects.create(
version=1,
data=b"This is just dummy data. PLEASE, DON'T UNPICKLE THIS !!",
name="Dummy model")
self.classifier_class = SKNaiveBayesClassifier
self.data = get_processed_data(csv_path)
def create_classifier_model(
version,
csv_path,
classifier_class=SKNaiveBayesClassifier,
confusion_matrix=True
):
"""
Create a new classifier object to save to the database
Parameters
----------
@classifier_class : Classifier class to use to creat model
@data : labeled data list [(text, classification), ...]
@version : version of the classifier model
"""
# check if version already exists
try:
ClassifierModel.objects.get(version=version)
raise Exception("Classifier version {} already exists".format(version))
except ClassifierModel.DoesNotExist:
pass
from helpers.deep import get_processed_data
data = get_processed_data(csv_path)
# get train, test data
train, test = create_train_test_data(data)
classifier = classifier_class.new(train)
accuracy = classifier.get_accuracy(test)
if confusion_matrix:
classifier.calculate_confusion_matrix(test)
pickle_data = pickle.dumps(classifier)
modelobj = ClassifierModel(
data=pickle_data,
accuracy=accuracy,
version=version,
name=classifier_class.__name__
)
testfilename = 'test_data_v-{}.pkl'.format(version)
filepath = 'model_test_datas/{}'.format(testfilename)
with open(filepath, 'wb') as f:
f.write(pickle.dumps(test))
modelobj.test_file_path = filepath
return modelobj
def main(*args, **kwargs):
if not kwargs.get('model_version'):
print("Version not provided. Provide it as --modelversion <version>")
return
csv_path = kwargs.get('path', '_playground/sample_data/processed_new_data.csv')
# TODO; check for model name
version = kwargs['model_version']
from helpers.deep import get_processed_data
# get data
data = get_processed_data(csv_path)
classifier_model = create_and_save_classifier_model(version, data)
print('Classifier {} created successfully with test data'.format(
classifier_model
))
def _get_model(self, version):
# first create classifier
csv_path = 'fixtures/processed_data_for_testing.csv'
data = get_processed_data(csv_path)
return create_classifier_model(version, data)
def setUp(self):
self.test_data = get_processed_data(
'fixtures/processed_data_for_testing.csv')
self.train, self.test = create_train_test_data(self.test_data)
self.classifier = SKNaiveBayesClassifier.new(self.train)
self.classifier.calculate_confusion_matrix(self.test)
logger = logging.getLogger('myapp')
hdlr = logging.FileHandler(logfilepath)
formatter = logging.Formatter('%(asctime)s %(levelname)s %(message)s')
hdlr.setFormatter(formatter)
logger.addHandler(hdlr)
logger.setLevel(logging.WARNING)
# logfile = open(logfilepath, 'w')
num_accuracy = []
try:
logger.info('.. GETTING DEEP DATA\n')
print('.. GETTING DEEP DATA\n')
deepdata = get_processed_data(
'_playground/sample_data/processed_sectors_subsectors.csv')
logger.info('.. SHUFFLING DEEP DATA\n')
print('.. SHUFFLING DEEP DATA\n')
random.shuffle(deepdata)
total = len(deepdata)
logger.info('.. INITIALIZING DATASETSIZE TO 500\n')
print('.. INITIALIZING DATASETSIZE TO 500\n')
dataset_num = 500
logger.info('.. SETTING SIZE INCREMENT TO 150\n')
print('.. SETTING SIZE INCREMENT TO 150\n')
increment = 150
# first create dir to store accuracy vs size data
logger.info('.. CREATING DIRECTORY `DEEP_DATA` FOR STORING DATA\n')
def test_create_train_test_data():
data = get_processed_data(csv_path)
train, test = create_train_test_data(data)
assert len(test) == int(len(data) / 4)
assert len(train) == int(3 * len(data) / 4)
def test_get_processed_data():
data = get_processed_data(csv_path)
assert type(data) == list, "The resulting data should be a list"
assert type(data[0]) == tuple, "Should be a tuple"
assert len(data[0]) == 2, "Tuple size should be 2"
def main(*args, **kwargs):
try:
logger.info('.. GETTING DEEP DATA\n')
print('.. GETTING DEEP DATA\n')
deepdata = get_processed_data(
'_playground/sample_data/processed_sectors_subsectors.csv')
logger.info('.. SHUFFLING DEEP DATA\n')
print('.. SHUFFLING DEEP DATA\n')
random.shuffle(deepdata)
total = len(deepdata)
logger.info('.. INITIALIZING DATASETSIZE TO 500\n')
print('.. INITIALIZING DATASETSIZE TO 500\n')
dataset_num = 500
logger.info('.. SETTING SIZE INCREMENT TO 150\n')
print('.. SETTING SIZE INCREMENT TO 150\n')
increment = 150
# first create dir to store accuracy vs size data
logger.info('.. CREATING DIRECTORY `DEEP_DATA` FOR STORING DATA\n')
print('.. CREATING DIRECTORY `DEEP_DATA` FOR STORING DATA\n')
dirpath = os.path.join(os.path.expanduser('~'), 'data_DEEPL')
subprocess.call(['mkdir', '-p', dirpath])
filepath = os.path.join(dirpath, 'accuracy_vs_size.txt')
logger.info('.. RUNNING LOOP')
print('.. RUNNING LOOP')
sectors_accuracies = {}
while dataset_num <= total:
random.shuffle(deepdata)
one_fourth = int(dataset_num / 4.0)
train = deepdata[:dataset_num][one_fourth:]
test = deepdata[:dataset_num][:one_fourth]
logger.info('.. dataset_num:{}\n'.format(dataset_num))
classifier = CLASSIFIER.new(train)
classifier.calculate_confusion_matrix(test)
# calculate accuracy for other
indices = classifier.confusion_matrix._indices
matrix = classifier.confusion_matrix._confusion
if not sectors_accuracies:
sectors_accuracies = {k: [] for k, v in indices.items()}
for k, v in indices:
total = sum(matrix[v])
correct = matrix[v][v]
sectors_accuracies[k].append(
[dataset_num, correct / float(total)])
accuracy = classifier.get_accuracy(test)
num_accuracy.append((dataset_num, accuracy))
logger.info('.. accuracy: {}\n'.format(accuracy))
print('.. accuracy: {}\n'.format(accuracy))
dataset_num += increment
# now plot
data = num_accuracy
x = list(map(lambda x: x[0], data))
y = list(map(lambda x: x[1], data))
print("$$$$$$$$$$$$$$$$$$$")
print(data)
print("$$$$$$$$$$$$$$$$$$$")
print(sectors_accuracies)
print("$$$$$$$$$$$$$$$$$$$")
fig = plt.figure(figsize=(15, 8))
plt.xticks([x for x in range(500, 28000, 1500)])
plt.xlabel('# of TRAINING SETS')
plt.ylabel('ACCURACY')
plt.grid(True)
plt.plot(x, y, 'k')
plt.savefig(str(datetime.datetime.now()) + ".png")
logger.info('.. DONE!!!')
except Exception as e:
import traceback
logger.info(traceback.format_exc())
print(traceback.format_exc())
logger.info('\n')
