Skip to content

Fix floating point issues (Issue #538) #589

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 9 commits into from
Dec 6, 2018
Merged

Fix floating point issues (Issue #538) #589

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
fd0fe3f
Merge pull request #579 from automl/development
mfeurer Nov 12, 2018
f0523fe
Pass train_size and test_size as integers (number of samples) instead…
Nov 30, 2018
84b28ea
Changed assigment of train samples for general use in every cv case
Dec 3, 2018
1ddf3a7
Added Unit Tests
Dec 3, 2018
267a328
Changed for simpler solution. Remove 'raveling' for multilabel cases
Dec 4, 2018
a046e8d
Fix PEP8 errors
Dec 4, 2018
971e9e7
Addition Fix PEP8 errors
Dec 4, 2018
9833d9c
Addition Fix PEP8 errors
Dec 4, 2018
e3241e1
Delete competition_c_functions.c
mfeurer Dec 6, 2018
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 9 additions & 7 deletions autosklearn/evaluation/train_evaluator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -540,30 +540,32 @@ def get_splitter(self, D):

return cv

y = D.data['Y_train'].ravel()
y = D.data['Y_train']
shuffle = self.resampling_strategy_args.get('shuffle', True)
train_size = 0.67
if self.resampling_strategy_args:
train_size = self.resampling_strategy_args.get('train_size',
train_size)
test_size = 1 - train_size
test_size = float("%.4f" % (1 - train_size))

if D.info['task'] in CLASSIFICATION_TASKS and \
D.info['task'] != MULTILABEL_CLASSIFICATION:

y = y.ravel()
if self.resampling_strategy in ['holdout',
'holdout-iterative-fit']:

if shuffle:
try:
cv = StratifiedShuffleSplit(n_splits=1,
train_size=train_size,
test_size=test_size,
random_state=1)
test_cv = copy.deepcopy(cv)
next(test_cv.split(y, y))
except ValueError as e:
if 'The least populated class in y has only' in e.args[0]:
cv = ShuffleSplit(n_splits=1, train_size=train_size,
test_size=test_size, random_state=1)
cv = ShuffleSplit(n_splits=1, test_size=test_size,
random_state=1)
else:
raise e
else:
Expand All @@ -588,8 +590,8 @@ def get_splitter(self, D):
'holdout-iterative-fit']:
# TODO shuffle not taken into account for this
if shuffle:
cv = ShuffleSplit(n_splits=1, train_size=train_size,
test_size=test_size, random_state=1)
cv = ShuffleSplit(n_splits=1, test_size=test_size,
random_state=1)
else:
tmp_train_size = int(np.floor(train_size * y.shape[0]))
test_fold = np.zeros(y.shape[0])
Expand Down
111 changes: 110 additions & 1 deletion test/test_evaluation/test_train_evaluator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,10 @@
eval_holdout, eval_iterative_holdout, eval_cv, eval_partial_cv
from autosklearn.util import backend
from autosklearn.util.pipeline import get_configuration_space
from autosklearn.constants import *
from autosklearn.constants import BINARY_CLASSIFICATION, \
MULTILABEL_CLASSIFICATION,\
MULTICLASS_CLASSIFICATION,\
REGRESSION
from autosklearn.metrics import accuracy, r2, f1_macro

this_directory = os.path.dirname(__file__)
Expand Down Expand Up @@ -1226,6 +1229,112 @@ def test_get_splitter_cv_object(self, te_mock):
next(cv.split(D.data['Y_train'], D.data['Y_train']
, groups=evaluator.resampling_strategy_args['groups']))

@unittest.mock.patch.object(TrainEvaluator, "__init__")
def test_holdout_split_size(self, te_mock):
te_mock.return_value = None
D = unittest.mock.Mock(spec=AbstractDataManager)
D.feat_type = []

evaluator = TrainEvaluator()
evaluator.resampling_strategy = 'holdout'

# Exact Ratio
D.data = dict(Y_train=np.array([0, 0, 0, 0, 0, 1, 1, 1, 1, 1]))
D.info = dict(task=BINARY_CLASSIFICATION)
evaluator.resampling_strategy_args = {'shuffle': True,
'train_size': 0.7}
cv = evaluator.get_splitter(D)

self.assertEqual(cv.get_n_splits(), 1)
train_samples, test_samples = next(cv.split(D.data['Y_train'],
D.data['Y_train']))
self.assertEqual(len(train_samples), 7)
self.assertEqual(len(test_samples), 3)

# No Shuffle
evaluator.resampling_strategy_args = {'shuffle': False,
'train_size': 0.7}
cv = evaluator.get_splitter(D)

self.assertEqual(cv.get_n_splits(), 1)
train_samples, test_samples = next(cv.split(D.data['Y_train'],
D.data['Y_train']))
self.assertEqual(len(train_samples), 7)
self.assertEqual(len(test_samples), 3)

# Rounded Ratio
D.data = dict(Y_train=np.array([0, 0, 0, 0, 0, 1, 1, 1, 1]))

evaluator.resampling_strategy_args = {'shuffle': True,
'train_size': 0.7}
cv = evaluator.get_splitter(D)

self.assertEqual(cv.get_n_splits(), 1)
train_samples, test_samples = next(cv.split(D.data['Y_train'],
D.data['Y_train']))
self.assertEqual(len(train_samples), 6)
self.assertEqual(len(test_samples), 3)

# Rounded Ratio No Shuffle
evaluator.resampling_strategy_args = {'shuffle': False,
'train_size': 0.7}
cv = evaluator.get_splitter(D)

self.assertEqual(cv.get_n_splits(), 1)
train_samples, test_samples = next(cv.split(D.data['Y_train'],
D.data['Y_train']))
self.assertEqual(len(train_samples), 6)
self.assertEqual(len(test_samples), 3)

# More data
evaluator.resampling_strategy_args = {'shuffle': True,
'train_size': 0.7}

D.data = dict(Y_train=np.zeros((900, 1)))
cv = evaluator.get_splitter(D)
self.assertEqual(cv.get_n_splits(), 1)
train_samples, test_samples = next(cv.split(D.data['Y_train'],
D.data['Y_train']))
self.assertEqual(len(train_samples), 630)
self.assertEqual(len(test_samples), 270)

evaluator.resampling_strategy_args = {'train_size': 0.752}
D.data = dict(Y_train=np.zeros((900, 1)))
cv = evaluator.get_splitter(D)
self.assertEqual(cv.get_n_splits(), 1)
train_samples, test_samples = next(cv.split(D.data['Y_train'],
D.data['Y_train']))
self.assertEqual(len(train_samples), 676)
self.assertEqual(len(test_samples), 224)

# Multilabel Exact Ratio
D.data = dict(Y_train=np.array([[0, 0], [0, 1], [1, 1], [1, 0], [1, 1],
[1, 1], [1, 1], [1, 0], [1, 1], [1, 1]]
))
D.info = dict(task=MULTILABEL_CLASSIFICATION)
evaluator.resampling_strategy_args = {'shuffle': True,
'train_size': 0.7}
cv = evaluator.get_splitter(D)

self.assertEqual(cv.get_n_splits(), 1)
train_samples, test_samples = next(cv.split(D.data['Y_train'],
D.data['Y_train']))
self.assertEqual(len(train_samples), 7)
self.assertEqual(len(test_samples), 3)

# Multilabel No Shuffle
D.data = dict(Y_train=np.array([[0, 0], [0, 1], [1, 1], [1, 0], [1, 1],
[1, 1], [1, 1], [1, 0], [1, 1]]))
evaluator.resampling_strategy_args = {'shuffle': False,
'train_size': 0.7}
cv = evaluator.get_splitter(D)

self.assertEqual(cv.get_n_splits(), 1)
train_samples, test_samples = next(cv.split(D.data['Y_train'],
D.data['Y_train']))
self.assertEqual(len(train_samples), 6)
self.assertEqual(len(test_samples), 3)


class FunctionsTest(unittest.TestCase):
def setUp(self):
Expand Down