8000 Merge branch 'master' of github.com:scikit-learn/scikit-learn into do… · donigian/scikit-learn@3bdcb15 · GitHub
[go: up one dir, main page]
Skip to content

Commit 3bdcb15

Browse files
donigiandonigian
committed
Merge branch 'master' of github.com:scikit-learn/scikit-learn into docs/donigian-update-contribution-guidelines
* 'master' of github.com:scikit-learn/scikit-learn: (23 commits) fixes scikit-learn#10031: fix attribute name and shape in documentation (scikit-learn#10033) [MRG+1] add changelog entry for fixed and merged PR scikit-learn#10005 issue scikit-learn#9633 (scikit-learn#10025) [MRG] Fix LogisticRegression see also should include LogisticRegressionCV(scikit-learn#9995) (scikit-learn#10022) [MRG + 1] Labels of clustering should start at 0 or -1 if noise (scikit-learn#10015) MAINT Remove redundancy in scikit-learn#9552 (scikit-learn#9573) [MRG+1] correct comparison in GaussianNB for 'priors' (scikit-learn#10005) [MRG + 1] ENH add check_inverse in FunctionTransformer (scikit-learn#9399) [MRG] FIX bug in nested set_params usage (scikit-learn#9999) [MRG+1] Fix LOF and Isolation benchmarks (scikit-learn#9798) [MRG + 1] Fix negative inputs checking in mean_squared_log_error (scikit-learn#9968) DOC Fix typo (scikit-learn#9996) DOC Fix typo: x axis -> y axis (scikit-learn#9985) improve example plot_forest_iris.py (scikit-learn#9989) [MRG+1] Deprecate pooling_func unused parameter in AgglomerativeClustering (scikit-learn#9875) DOC update news DOC Fix three typos in manifold documentation (scikit-learn#9990) DOC add missing dot in docstring DOC Add what's new for 0.19.1 (scikit-learn#9983) Improve readability of outlier detection example. (scikit-learn#9973) DOC: Fixed typo (scikit-learn#9977) ...
2 parents e8e2b39 + 5cf3fb7 commit 3bdcb15

37 files changed

+445
-121
lines changed

‎benchmarks/bench_isolation_forest.py

Lines changed: 26 additions & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -3,6 +3,17 @@
33
IsolationForest benchmark
44
==========================================
55
A test of IsolationForest on classical anomaly detection datasets.
6+
7+
The benchmark is run as f F438 ollows:
8+
1. The dataset is randomly split into a training set and a test set, both
9+
assumed to contain outliers.
10+
2. Isolation Forest is trained on the training set.
11+
3. The ROC curve is computed on the test set using the knowledge of the labels.
12+
13+
Note that the smtp dataset contains a very small proportion of outliers.
14+
Therefore, depending on the seed of the random number generator, randomly
15+
splitting the data set might lead to a test set containing no outliers. In this
16+
case a warning is raised when computing the ROC curve.
617
"""
718

819
from time import time
@@ -12,7 +23,7 @@
1223
from sklearn.ensemble import IsolationForest
1324
from sklearn.metrics import roc_curve, auc
1425
from sklearn.datasets import fetch_kddcup99, fetch_covtype, fetch_mldata
15-
from sklearn.preprocessing import MultiLabelBinarizer
26+
from sklearn.preprocessing import LabelBinarizer
1627
from sklearn.utils import shuffle as sh
1728

1829
print(__doc__)
@@ -30,15 +41,14 @@ def print_outlier_ratio(y):
3041
print("----- Outlier ratio: %.5f" % (np.min(cnt) / len(y)))
3142

3243

33-
np.random.seed(1)
44+
random_state = 1
3445
fig_roc, ax_roc = plt.subplots(1, 1, figsize=(8, 5))
3546

3647
# Set this to true for plotting score histograms for each dataset:
3748
with_decision_function_histograms = False
3849

39-
# Removed the shuttle dataset because as of 2017-03-23 mldata.org is down:
40-
# datasets = ['http', 'smtp', 'SA', 'SF', 'shuttle', 'forestcover']
41-
datasets = ['http', 'smtp', 'SA', 'SF', 'forestcover']
50+
# datasets available = ['http', 'smtp', 'SA', 'SF', 'shuttle', 'forestcover']
51+
datasets = ['http', 'smtp', 'SA', 'SF', 'shuttle', 'forestcover']
4252

4353
# Loop over all datasets for fitting and scoring the estimator:
4454
for dat in datasets:
@@ -47,15 +57,16 @@ def print_outlier_ratio(y):
4757
print('====== %s ======' % dat)
4858
print('--- Fetching data...')
4959
if dat in ['http', 'smtp', 'SF', 'SA']:
50-
dataset = fetch_kddcup99(subset=dat, shuffle=True, percent10=True)
60+
dataset = fetch_kddcup99(subset=dat, shuffle=True,
61+
percent10=True, random_state=random_state)
5162
X = dataset.data
5263
y = dataset.target
5364

5465
if dat == 'shuttle':
5566
dataset = fetch_mldata('shuttle')
5667
X = dataset.data
5768
y = dataset.target
58-
X, y = sh(X, y)
69+
X, y = sh(X, y, random_state=random_state)
5970
# we remove data with label 4
6071
# normal data are then those of class 1
6172
s = (y != 4)
@@ -65,7 +76,7 @@ def print_outlier_ratio(y):
6576
print('----- ')
6677

6778
if dat == 'forestcover':
68-
dataset = fetch_covtype(shuffle=True)
79+
dataset = fetch_covtype(shuffle=True, random_state=random_state)
6980
X = dataset.data
7081
y = dataset.target
7182
# normal data are those with attribute 2
@@ -79,17 +90,17 @@ def print_outlier_ratio(y):
7990
print('--- Vectorizing data...')
8091

8192
if dat == 'SF':
82-
lb = MultiLabelBinarizer()
83-
x1 = lb.fit_transform(X[:, 1])
93+
lb = LabelBinarizer()
94+
x1 = lb.fit_transform(X[:, 1].astype(str))
8495
X = np.c_[X[:, :1], x1, X[:, 2:]]
8596
y = (y != b'normal.').astype(int)
8697
print_outlier_ratio(y)
8798

8899
if dat == 'SA':
89-
lb = MultiLabelBinarizer()
90-
x1 = lb.fit_transform(X[:, 1])
91-
x2 = lb.fit_transform(X[:, 2])
92-
x3 = lb.fit_transform(X[:, 3])
100+
lb = LabelBinarizer()
101+
x1 = lb.fit_transform(X[:, 1].astype(str))
102+
x2 = lb.fit_transform(X[:, 2].astype(str))
103+
x3 = lb.fit_transform(X[:, 3].astype(str))
93104
X = np.c_[X[:, :1], x1, x2, x3, X[:, 4:]]
94105
y = (y != b'normal.').astype(int)
95106
print_outlier_ratio(y)
@@ -108,7 +119,7 @@ def print_outlier_ratio(y):
108119
y_test = y[n_samples_train:]
109120

110121
print('--- Fitting the IsolationForest estimator...')
111-
model = IsolationForest(n_jobs=-1)
122+
model = IsolationForest(n_jobs=-1, random_state=random_state)
112123
tstart = time()
113124
model.fit(X_train)
114125
fit_time = time() - tstart

‎benchmarks/bench_lof.py

Lines changed: 28 additions & 41 deletions
Original file line numberDiff line numberDiff line change
@@ -5,6 +5,16 @@
55
66
A test of LocalOutlierFactor on classical anomaly detection datasets.
77
8+
Note that LocalOutlierFactor is not meant to predict on a test set and its
9+
performance is assessed in an outlier detection context:
10+
1. The model is trained on the whole dataset which is assumed to contain
11+
outliers.
12+
2. The ROC curve is computed on the same dataset using the knowledge of the
13+
labels.
14+
In this context there is no need to shuffle the dataset because the model
15+
is trained and tested on the whole dataset. The randomness of this benchmark
16+
is only caused by the random selection of anomalies in the SA dataset.
17+
818
"""
919

1020
from time import time
@@ -14,31 +24,28 @@
1424
from sklearn.metrics import roc_curve, auc
1525
from sklearn.datasets import fetch_kddcup99, fetch_covtype, fetch_mldata
1626
from sklearn.preprocessing import LabelBinarizer
17-
from sklearn.utils import shuffle as sh
1827

1928
print(__doc__)
2029

21-
np.random.seed(2)
30+
random_state = 2 # to control the random selection of anomalies in SA
2231

2332
# datasets available: ['http', 'smtp', 'SA', 'SF', 'shuttle', 'forestcover']
24-
datasets = ['shuttle']
25-
26-
novelty_detection = True # if False, training set polluted by outliers
33+
datasets = ['http', 'smtp', 'SA', 'SF', 'shuttle', 'forestcover']
2734

35+
plt.figure()
2836
for dataset_name in datasets:
2937
# loading and vectorization
3038
print('loading data')
3139
if dataset_name in ['http', 'smtp', 'SA', 'SF']:
32-
dataset = fetch_kddcup99(subset=dataset_name, shuffle=True,
33-
percent10=False)
40+
dataset = fetch_kddcup99(subset=dataset_name, percent10=True,
41+
random_state=random_state)
3442
X = dataset.data
3543
y = dataset.target
3644

3745
if dataset_name == 'shuttle':
3846
dataset = fetch_mldata('shuttle')
3947
X = dataset.data
4048
y = dataset.target
41-
X, y = sh(X, y)
4249
# we remove data with label 4
4350
# normal data are then those of class 1
4451
s = (y != 4)
@@ -47,7 +54,7 @@
4754
y = (y != 1).astype(int)
4855

4956
if dataset_name == 'forestcover':
50-
dataset = fetch_covtype(shuffle=True)
57+
dataset = fetch_covtype()
5158
X = dataset.data
5259
y = dataset.target
5360
# normal data are those with attribute 2
@@ -61,54 +68,34 @@
6168

6269
if dataset_name == 'SF':
6370
lb = LabelBinarizer()
64-
lb.fit(X[:, 1])
65-
x1 = lb.transform(X[:, 1])
71+
x1 = lb.fit_transform(X[:, 1].astype(str))
6672
X = np.c_[X[:, :1], x1, X[:, 2:]]
67-
y = (y != 'normal.').astype(int)
73+
y = (y != b'normal.').astype(int)
6874

6975
if dataset_name == 'SA':
7076
lb = LabelBinarizer()
71-
lb.fit(X[:, 1])
72-
x1 = lb.transform(X[:, 1])
73-
lb.fit(X[:, 2])
74-
x2 = lb.transform(X[:, 2])
75-
lb.fit(X[:, 3])
76-
x3 = lb.transform(X[:, 3])
77+
x1 = lb.fit_transform(X[:, 1].astype(str))
78+
x2 = lb.fit_transform(X[:, 2].astype(str))
79+
x3 = lb.fit_transform(X[:, 3].astype(str))
7780
X = np.c_[X[:, :1], x1, x2, x3, X[:, 4:]]
78-
y = (y != 'normal.').astype(int)
81+
y = (y != b'normal.').astype(int)
7982

8083
if dataset_name == 'http' or dataset_name == 'smtp':
81-
y = (y != 'normal.').astype(int)
82-
83-
n_samples, n_features = np.shape(X)
84-
n_samples_train = n_samples // 2
85-
n_samples_test = n_samples - n_samples_train
84+
y = (y != b'normal.').astype(int)
8685

8786
X = X.astype(float)
88-
X_train = X[:n_samples_train, :]
89-
X_test = X[n_samples_train:, :]
90-
y_train = y[:n_samples_train]
91-
y_test = y[n_samples_train:]
92-
93-
if novelty_detection:
94-
X_train = X_train[y_train == 0]
95-
y_train = y_train[y_train == 0]
9687

9788
print('LocalOutlierFactor processing...')
9889
model = LocalOutlierFactor(n_neighbors=20)
9990
tstart = time()
100-
model.fit(X_train)
91+
model.fit(X)
10192
fit_time = time() - tstart
102-
tstart = time()
103-
104-
scoring = -model.decision_function(X_test) # the lower, the more normal
105-
predict_time = time() - tstart
106-
fpr, tpr, thresholds = roc_curve(y_test, scoring)
93+
scoring = -model.negative_outlier_factor_ # the lower, the more normal
94+
fpr, tpr, thresholds = roc_curve(y, scoring)
10795
AUC = auc(fpr, tpr)
10896
plt.plot(fpr, tpr, lw=1,
109-
label=('ROC for %s (area = %0.3f, train-time: %0.2fs,'
110-
'test-time: %0.2fs)' % (dataset_name, AUC, fit_time,
111-
predict_time)))
97+
label=('ROC for %s (area = %0.3f, train-time: %0.2fs)'
98+
% (dataset_name, AUC, fit_time)))
11299

113100
plt.xlim([-0.05, 1.05])
114101
plt.ylim([-0.05, 1.05])

‎doc/datasets/index.rst

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -321,6 +321,7 @@ writing data in that format.
321321
labeled_faces
322322
covtype
323323
rcv1
324+
kddcup99
324325

325326

326327
.. include:: olivetti_faces.rst
@@ -335,6 +336,8 @@ writing data in that format.
335336

336337
.. include:: rcv1.rst
337338

339+
.. include:: kddcup99.rst
340+
338341
.. _boston_house_prices:
339342

340343
.. include:: ../../sklearn/datasets/descr/boston_house_prices.rst

‎doc/index.rst

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -207,6 +207,8 @@
207207
<li><em>On-going development:</em>
208208
<a href="/dev/whats_new.html"><em>What's new</em> (Changelog)</a>
209209
</li>
210+
<li><em>October 2017.</em> scikit-learn 0.19.1 is available for download (<a href="whats_new.html#version-0-19">Changelog</a>).
211+
</li>
210212
<li><em>July 2017.</em> scikit-learn 0.19.0 is available for download (<a href="whats_new/v0.19.html#version-0-19">Changelog</a>).
211213
</li>
212214
<li><em>June 2017.</em> scikit-learn 0.18.2 is available for download (<a href="whats_new/v0.18.html#version-0-18-2">Changelog</a>).

‎doc/modules/manifold.rst

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -533,7 +533,7 @@ the quality of the resulting embedding:
533533
* maximum number of iterations
534534
* angle (not used in the exact method)
535535

536-
The perplexity is defined as :math:`k=2^(S)` where :math:`S` is the Shannon
536+
The perplexity is defined as :math:`k=2^{(S)}` where :math:`S` is the Shannon
537537
entropy of the conditional probability distribution. The perplexity of a
538538
:math:`k`-sided die is :math:`k`, so that :math:`k` is effectively the number of
539539
nearest neighbors t-SNE considers when generating the conditional probabilities.
@@ -598,8 +598,8 @@ where label regions largely overlap. This is a strong clue that this data can
598598
be well separated by non linear methods that focus on the local structure (e.g.
599599
an SVM with a Gaussian RBF kernel). However, failing to visualize well
600600
separated homogeneously labeled groups with t-SNE in 2D does not necessarily
601-
implie that the data cannot be correctly classified by a supervised model. It
602-
might be the case that 2 dimensions are not enough low to accurately represents
601+
imply that the data cannot be correctly classified by a supervised model. It
602+
might be the case that 2 dimensions are not low enough to accurately represents
603603
the internal structure of the data.
604604

605605

‎doc/modules/preprocessing.rst

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -610,6 +610,15 @@ a transformer that applies a log transformation in a pipeline, do::
610610
array([[ 0. , 0.69314718],
611611
[ 1.09861229, 1.38629436]])
612612

613+
You can ensure that ``func`` and ``inverse_func`` are the inverse of each other
614+
by setting ``check_inverse=True`` and calling ``fit`` before
615+
``transform``. Please note that a warning is raised and can be turned into an
616+
error with a ``filterwarnings``::
617+
618+
>>> import warnings
619+
>>> warnings.filterwarnings("error", message=".*check_inverse*.",
620+
... category=UserWarning, append=False)
621+
613622
For a full code example that demonstrates using a :class:`FunctionTransformer`
614623
to do custom feature selection,
615624
see :ref:`sphx_glr_auto_examples_preprocessing_plot_function_transformer.py`

0 commit comments

Comments
 (0)
0