def reset(self, params, rep):
name = params['dataset']
print params
random.seed(abs(hash(str(params))))
if name.startswith('toy'):
data_dims, cone_dims = [int(x) for x in
name.split('-')[1:]]
self.dimensions = [cone_dims]
self.dataset = make_data(data_dims, cone_dims)
elif name.startswith('wn'):
self.dimensions = params['dimensions']
self.dataset = SvmlightDataset(
load_svmlight_file('../../../Documents/conewordnetdata/data-nouns-deps-mi/' + name + '.mat'))
print self.dataset.target.shape
print self.dataset.data.shape
else:
self.dimensions = params['dimensions']
self.dataset = fetch_mldata(name)
# Ensure that the data is always shuffled the same way:
# seed RNG on data itself
seed = int(hashlib.sha1(self.dataset.data).hexdigest()[:7], 16)
# print len(self.dataset.data), self.dataset.target.shape[0]
# shuffled_data, shuffled_target = utils.shuffle(
# self.dataset.data, self.dataset.target, random_state = seed)
# StratifiedKFold is deterministic
self.cv = KFold(k = params['repetitions'], n = self.dataset.target.shape[0],
shuffle = True, random_state = seed)
train, test = list(self.cv)[rep]
print len(train), len(test)
self.X_train = self.dataset.data[train]
self.X_test = self.dataset.data[test]
self.y_train = self.dataset.target[train]
self.y_test = self.dataset.target[test]
def generator(data_dims, cone_dims, num_instances=1000):
return make_data(
data_dims, cone_dims, size=num_instances, epsilon=epsilon)
def generateNoisyTestData(self, data_dims, cone_dims, num_instances=1000):
return make_data(data_dims, cone_dims, size=num_instances, noise=0.1)
