diff --git a/scikits/learn/__init__.py b/scikits/learn/__init__.py
index df11355d259d7564752d73affb7b50d0775d21da..3d2cce01fdd3720a152f00229f4fe60dd007e84a 100644
--- a/scikits/learn/__init__.py
+++ b/scikits/learn/__init__.py
@@ -18,9 +18,11 @@ from .base import clone
try:
from numpy.testing import nosetester
+
class NoseTester(nosetester.NoseTester):
""" Subclass numpy's NoseTester to add doctests by default
"""
+
def test(self, label='fast', verbose=1, extra_argv=['--exe'],
doctests=True, coverage=False):
return super(NoseTester, self).test(label=label, verbose=verbose,
@@ -33,9 +35,8 @@ except:
pass
-__all__ = ['cross_val', 'ball_tree', 'cluster', 'covariance', 'datasets', 'gmm',
- 'linear_model', 'logistic', 'lda', 'metrics', 'svm', 'features', 'clone',
- 'test', 'gaussian_process']
+__all__ = ['cross_val', 'ball_tree', 'cluster', 'covariance', 'datasets',
+ 'gmm', 'linear_model', 'logistic', 'lda', 'metrics', 'svm',
+ 'features', 'clone', 'metrics', 'test', 'gaussian_process']
__version__ = '0.6.git'
-
diff --git a/scikits/learn/cluster/mean_shift_.py b/scikits/learn/cluster/mean_shift_.py
index 1b9934d09880c6fbd13162fcc569dd90d59ef129..8257942a21c9732398d028cbf1ca35dcc1039d11 100644
--- a/scikits/learn/cluster/mean_shift_.py
+++ b/scikits/learn/cluster/mean_shift_.py
@@ -9,38 +9,7 @@ from math import floor
import numpy as np
from ..base import BaseEstimator
-
-def euclidian_distances(X, Y=None):
- """
- Considering the rows of X (and Y=X) as vectors, compute the
- distance matrix between each pair of vector
-
- Parameters
- ----------
- X, array of shape (n_samples_1, n_features)
-
- Y, array of shape (n_samples_2, n_features), default None
- if Y is None, then Y=X is used instead
-
- Returns
- -------
- distances, array of shape (n_samples_1, n_samples_2)
- """
- if Y is None:
- Y = X
- if X.shape[1] != Y.shape[1]:
- raise ValueError, "incompatible dimension for X and Y matrices"
-
- XX = np.sum(X * X, axis=1)[:,np.newaxis]
- if Y is None:
- YY = XX.T
- else:
- YY = np.sum(Y * Y, axis=1)[np.newaxis,:]
- distances = XX + YY # Using broadcasting
- distances -= 2 * np.dot(X, Y.T)
- distances = np.maximum(distances, 0)
- distances = np.sqrt(distances)
- return distances
+from ..metrics.pairwise import euclidian_distances
def estimate_bandwidth(X, quantile=0.3):
@@ -101,7 +70,7 @@ def mean_shift(X, bandwidth=None):
n_clusters = 0
bandwidth_squared = bandwidth**2
points_idx_init = np.arange(n_points)
- stop_thresh = 1e-3*bandwidth # when mean has converged
+ stop_thresh = 1e-3*bandwidth # when mean has converged
cluster_centers = [] # center of clusters
# track if a points been seen already
been_visited_flag = np.zeros(n_points, dtype=np.bool)
@@ -114,12 +83,12 @@ def mean_shift(X, bandwidth=None):
while n_points_init:
# pick a random seed point
- tmp_index = random_state.randint(n_points_init)
+ tmp_index = random_state.randint(n_points_init)
# use this point as start of mean
- start_idx = points_idx_init[tmp_index]
- my_mean = X[start_idx, :] # intilize mean to this points location
+ start_idx = points_idx_init[tmp_index]
+ my_mean = X[start_idx, :] # intilize mean to this points location
# points that will get added to this cluster
- my_members = np.zeros(n_points, dtype=np.bool)
+ my_members = np.zeros(n_points, dtype=np.bool)
# used to resolve conflicts on cluster membership
this_cluster_votes = np.zeros(n_points, dtype=np.uint16)
@@ -133,10 +102,10 @@ def mean_shift(X, bandwidth=None):
# add a vote for all the in points belonging to this cluster
this_cluster_votes[in_idx] += 1
- my_old_mean = my_mean # save the old mean
- my_mean = np.mean(X[in_idx,:], axis=0) # compute the new mean
+ my_old_mean = my_mean # save the old mean
+ my_mean = np.mean(X[in_idx, :], axis=0) # compute the new mean
# add any point within bandwidth to the cluster
- my_members = np.logical_or(my_members, in_idx)
+ my_members = np.logical_or(my_members, in_idx)
# mark that these points have been visited
been_visited_flag[my_members] = True
@@ -169,7 +138,7 @@ def mean_shift(X, bandwidth=None):
# we can initialize with any of the points not yet visited
points_idx_init = np.where(been_visited_flag == False)[0]
- n_points_init = points_idx_init.size # number of active points in set
+ n_points_init = points_idx_init.size # number of active points in set
# a point belongs to the cluster with the most votes
labels = np.argmax(cluster_votes, axis=0)
@@ -177,7 +146,8 @@ def mean_shift(X, bandwidth=None):
return cluster_centers, labels
-################################################################################
+##############################################################################
+
class MeanShift(BaseEstimator):
"""MeanShift clustering
@@ -233,5 +203,3 @@ class MeanShift(BaseEstimator):
self._set_params(**params)
self.cluster_centers_, self.labels_ = mean_shift(X, self.bandwidth)
return self
-
-
diff --git a/scikits/learn/metrics/__init__.py b/scikits/learn/metrics/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..78e3131f9e013868b99e7254cbd8be32c7cca7af
--- /dev/null
+++ b/scikits/learn/metrics/__init__.py
@@ -0,0 +1,10 @@
+"""
+Metrics module with score functions, performance metrics and
+pairwise metrics or distances computation
+"""
+
+from .metrics import confusion_matrix, roc_curve, auc, precision_score, \
+ recall_score, fbeta_score, f1_score, \
+ precision_recall_fscore_support, classification_report, \
+ precision_recall_curve, explained_variance_score, r2_score, \
+ zero_one, mean_square_error
diff --git a/scikits/learn/metrics.py b/scikits/learn/metrics/metrics.py
similarity index 98%
rename from scikits/learn/metrics.py
rename to scikits/learn/metrics/metrics.py
index c2e646146201dcc17f33b845791e79a3e067187f..6d8100107b210830d0eaafa278eb325a572adee9 100644
--- a/scikits/learn/metrics.py
+++ b/scikits/learn/metrics/metrics.py
@@ -166,7 +166,7 @@ def precision_score(y_true, y_pred, pos_label=1):
precision of the positive class in binary classification or
weighted avergage of the precision of each class for the
multiclass task
-
+
"""
p, _, _, s = precision_recall_fscore_support(y_true, y_pred)
if p.shape[0] == 2:
@@ -417,7 +417,8 @@ def classification_report(y_true, y_pred, labels=None, class_names=None):
report = fmt % tuple(headers)
report += '\n'
- p, r, f1, s = precision_recall_fscore_support(y_true, y_pred, labels=labels)
+ p, r, f1, s = precision_recall_fscore_support(y_true, y_pred,
+ labels=labels)
for i, label in enumerate(labels):
values = [class_names[i]]
for v in (p[i], r[i], f1[i]):
@@ -512,7 +513,7 @@ def explained_variance_score(y_true, y_pred):
y_pred : array-like
"""
return 1 - np.var(y_true - y_pred) / np.var(y_true)
-
+
def r2_score(y_true, y_pred):
"""R^2 (coefficient of determination) regression score function
@@ -529,7 +530,8 @@ def r2_score(y_true, y_pred):
y_pred : array-like
"""
- return 1 - ((y_true - y_pred)**2).sum() / ((y_true - y_true.mean())**2).sum()
+ return 1 - (((y_true - y_pred)**2).sum() /
+ ((y_true - y_true.mean())**2).sum())
@@ -539,7 +541,7 @@ def r2_score(y_true, y_pred):
def zero_one(y_true, y_pred):
"""Zero-One classification loss
- Positive integer (number of misclassifications). The best performance
+ Positive integer (number of misclassifications). The best performance
is 0.
Return the number of errors
@@ -575,6 +577,3 @@ def mean_square_error(y_true, y_pred):
loss : float
"""
return np.linalg.norm(y_pred - y_true) ** 2
-
-
-
diff --git a/scikits/learn/metrics/pairwise.py b/scikits/learn/metrics/pairwise.py
new file mode 100644
index 0000000000000000000000000000000000000000..ef32cae79729271a17b0d611977dd4dd17d45aa2
--- /dev/null
+++ b/scikits/learn/metrics/pairwise.py
@@ -0,0 +1,44 @@
+"""Utilities to evaluate pairwise distances or metrics between 2
+sets of points.
+
+"""
+
+# Authors: Alexandre Gramfort <alexandre.gramfort@inria.fr>
+# License: BSD Style.
+
+import numpy as np
+
+
+def euclidian_distances(X, Y=None):
+ """
+ Considering the rows of X (and Y=X) as vectors, compute the
+ distance matrix between each pair of vector
+
+ Parameters
+ ----------
+ X, array of shape (n_samples_1, n_features)
+
+ Y, array of shape (n_samples_2, n_features), default None
+ if Y is None, then Y=X is used instead
+
+ Returns
+ -------
+ distances, array of shape (n_samples_1, n_samples_2)
+ """
+ X = np.asanyarray(X)
+ Y = np.asanyarray(Y)
+ if Y is None:
+ Y = X
+ if X.shape[1] != Y.shape[1]:
+ raise ValueError, "incompatible dimension for X and Y matrices"
+
+ XX = np.sum(X * X, axis=1)[:, np.newaxis]
+ if Y is None:
+ YY = XX.T
+ else:
+ YY = np.sum(Y * Y, axis=1)[np.newaxis, :]
+ distances = XX + YY # Using broadcasting
+ distances -= 2 * np.dot(X, Y.T)
+ distances = np.maximum(distances, 0)
+ distances = np.sqrt(distances)
+ return distances
diff --git a/scikits/learn/metrics/tests/__init__.py b/scikits/learn/metrics/tests/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/scikits/learn/tests/test_metrics.py b/scikits/learn/metrics/tests/test_metrics.py
similarity index 99%
rename from scikits/learn/tests/test_metrics.py
rename to scikits/learn/metrics/tests/test_metrics.py
index f3c8896651d5c508ac54929819c1a821ae622b90..212e4b2a1a8f4eaf21b80230e38f66fc86c31f37 100644
--- a/scikits/learn/tests/test_metrics.py
+++ b/scikits/learn/metrics/tests/test_metrics.py
@@ -7,8 +7,8 @@ from numpy.testing import assert_array_almost_equal
from numpy.testing import assert_array_equal
from numpy.testing import assert_equal, assert_almost_equal
-from .. import datasets
-from .. import svm
+from ... import datasets
+from ... import svm
from ..metrics import auc
from ..metrics import classification_report
from ..metrics import confusion_matrix
diff --git a/scikits/learn/metrics/tests/test_pairwise.py b/scikits/learn/metrics/tests/test_pairwise.py
new file mode 100644
index 0000000000000000000000000000000000000000..1d8b1dc54eb8d033517e61536b116691ffe8ae87
--- /dev/null
+++ b/scikits/learn/metrics/tests/test_pairwise.py
@@ -0,0 +1,11 @@
+from numpy.testing import assert_array_almost_equal
+
+from ..pairwise import euclidian_distances
+
+
+def test_euclidian_distances():
+ """Check that the pairwise euclidian distances computation"""
+ X = [[0]]
+ Y = [[1], [2]]
+ D = euclidian_distances(X, Y)
+ assert_array_almost_equal(D, [[1., 2.]])
diff --git a/scikits/learn/setup.py b/scikits/learn/setup.py
index b50541f2ae919a23b494ac73d1fc6537421c7e9e..b2e940a486a73e66103b892ca5c98c39f2016ef2 100644
--- a/scikits/learn/setup.py
+++ b/scikits/learn/setup.py
@@ -3,6 +3,7 @@ import warnings
import numpy
import sys
+
def configuration(parent_package='', top_path=None):
from numpy.distutils.misc_util import Configuration
from numpy.distutils.system_info import get_info, BlasNotFoundError
@@ -26,14 +27,15 @@ def configuration(parent_package='', top_path=None):
config.add_subpackage('externals')
config.add_subpackage('gaussian_process')
config.add_subpackage('gaussian_process/tests')
+ config.add_subpackage('metrics')
+ config.add_subpackage('metrics/tests')
# some libs needs cblas, fortran-compiled BLAS will not be sufficient
blas_info = get_info('blas_opt', 0)
if (not blas_info) or (
- ('NO_ATLAS_INFO', 1) in blas_info.get('define_macros', [])) :
+ ('NO_ATLAS_INFO', 1) in blas_info.get('define_macros', [])):
config.add_library('cblas',
- sources=[join('src', 'cblas', '*.c')]
- )
+ sources=[join('src', 'cblas', '*.c')])
cblas_libs = ['cblas']
blas_info.pop('libraries', None)
warnings.warn(BlasNotFoundError.__doc__)
@@ -43,8 +45,7 @@ def configuration(parent_package='', top_path=None):
config.add_extension('ball_tree',
sources=[join('src', 'BallTree.cpp')],
- include_dirs=[numpy.get_include()]
- )
+ include_dirs=[numpy.get_include()])
# the following packages depend on cblas, so they have to be build