def generate_samples(partial_policy, start_index = 0):
"""
Generates all possible ways to complete a partial policy.
This is a recursive method meant to be used internally only.
For the public use of this call generate_training_set
:partial_policy: The policy that is currently in the process of construction
:start_index: Used to manage progress to avoid repetitions
"""
list = [] # policies
indexedOptions = policy_model.get_ranges().keys()
for i in range(start_index, len(indexedOptions)): # search for first
policy = indexedOptions[i]
# for policy in indexedOptions:
if not policy in partial_policy:
for value in policy_model.get_ranges()[policy]:
new_partial = partial_policy.copy()
new_partial[policy] = value
complete_new = classifier_sklearn.generate_samples(new_partial, start_index=i)
list.extend(complete_new)
return list # stop loop when first value is found and recursion for it is done
return [partial_policy]
def train_engine(self,
risk,
employee,
location,
device,
type="classifier"):
data = self.load_csv_files(risk, employee, location, device, type=type)
limit = len(policy_model.get_ranges())
train_data = data[:, 0:
limit] # first several columns represent the data dimension
train_result = data[:,
limit] # result columns are ones after data dimensions
# TODO: implement weighting
# weighted_train_data = self.weight(train_data)
# for explanation of the following parameters please see
# http://scikit-learn.org/stable/modules/svm.html#tips-on-practical-use
params = {'kernel': 'rbf', 'cache_size': 1000, 'C': 0.2, 'gamma': 0.5}
eng = engine.get_engine(type=type)
# TODO: implement weighting
# return eng.get_model(params).fit(weighted_train_data, train_result)
return eng.get_model(params).fit(train_data, train_result)
def __init__(self):
"""
Initializes all implicit models.
Currently there is one model per risk,
but should turn into one model per risk per environmental configuration
:param user_id:
:param sync_date:
"""
self.incidents_models = {}
self.risks = []
limit = len(policy_model.get_ranges())
general = numpy.genfromtxt('static/data/pw-train-generated-general.csv', delimiter=',')
for filename in glob.glob('static/data/pw-train-generated-risk-*.csv'):
risk = filename[36:-4] # take actual name
self.risks.append(risk)
# data = genfromtxt('static/data/pw-train-estimator-risk-' + risk + '.csv', delimiter=',')
data = numpy.genfromtxt(filename, delimiter=',')
data = numpy.concatenate((data, general)) # add positive cases that need to contrast negative ones
train_data = data[:, 0:limit] # first several columns represent the data dimension
train_result = data[:, limit] # result columns are ones after data dimensions
self.incidents_models[risk] = svm.SVC().fit(train_data, train_result)
