def test_linearsvx_loss_penalty_deprecations():
X, y = [[0.0], [1.0]], [0, 1]
msg = ("loss='%s' has been deprecated in favor of "
"loss='%s' as of 0.16. Backward compatibility"
" for the %s will be removed in %s")
# LinearSVC
# loss l1 --> hinge
assert_warns_message(DeprecationWarning,
msg % ("l1", "hinge", "loss='l1'", "0.23"),
svm.LinearSVC(loss="l1").fit, X, y)
# loss l2 --> squared_hinge
assert_warns_message(DeprecationWarning,
msg % ("l2", "squared_hinge", "loss='l2'", "0.23"),
svm.LinearSVC(loss="l2").fit, X, y)
# LinearSVR
# loss l1 --> epsilon_insensitive
assert_warns_message(
DeprecationWarning,
msg % ("l1", "epsilon_insensitive", "loss='l1'", "0.23"),
svm.LinearSVR(loss="l1").fit, X, y)
# loss l2 --> squared_epsilon_insensitive
assert_warns_message(
DeprecationWarning,
msg % ("l2", "squared_epsilon_insensitive", "loss='l2'", "0.23"),
svm.LinearSVR(loss="l2").fit, X, y)
def test_linearsvc():
# Test basic routines using LinearSVC
clf = svm.LinearSVC(random_state=0).fit(X, Y)
# by default should have intercept
assert clf.fit_intercept
assert_array_equal(clf.predict(T), true_result)
assert_array_almost_equal(clf.intercept_, [0], decimal=3)
# the same with l1 penalty
clf = svm.LinearSVC(penalty='l1',
loss='squared_hinge',
dual=False,
random_state=0).fit(X, Y)
assert_array_equal(clf.predict(T), true_result)
# l2 penalty with dual formulation
clf = svm.LinearSVC(penalty='l2', dual=True, random_state=0).fit(X, Y)
assert_array_equal(clf.predict(T), true_result)
# l2 penalty, l1 loss
clf = svm.LinearSVC(penalty='l2', loss='hinge', dual=True, random_state=0)
clf.fit(X, Y)
assert_array_equal(clf.predict(T), true_result)
# test also decision function
dec = clf.decision_function(T)
res = (dec > 0).astype(np.int) + 1
assert_array_equal(res, true_result)
def test_linearsvc_parameters():
# Test possible parameter combinations in LinearSVC
# Generate list of possible parameter combinations
losses = ['hinge', 'squared_hinge', 'logistic_regression', 'foo']
penalties, duals = ['l1', 'l2', 'bar'], [True, False]
X, y = make_classification(n_samples=5, n_features=5)
for loss, penalty, dual in itertools.product(losses, penalties, duals):
clf = svm.LinearSVC(penalty=penalty, loss=loss, dual=dual)
if ((loss, penalty) == ('hinge', 'l1')
or (loss, penalty, dual) == ('hinge', 'l2', False)
or (penalty, dual) == ('l1', True) or loss == 'foo'
or penalty == 'bar'):
with pytest.raises(ValueError,
match="Unsupported set of "
"arguments.*penalty='%s.*loss='%s.*dual=%s" %
(penalty, loss, dual)):
clf.fit(X, y)
else:
clf.fit(X, y)
# Incorrect loss value - test if explicit error message is raised
with pytest.raises(ValueError, match=".*loss='l3' is not supported.*"):
svm.LinearSVC(loss="l3").fit(X, y)
def test_linearsvc_fit_sampleweight():
# check correct result when sample_weight is 1
n_samples = len(X)
unit_weight = np.ones(n_samples)
clf = svm.LinearSVC(random_state=0).fit(X, Y)
clf_unitweight = svm.LinearSVC(random_state=0).\
fit(X, Y, sample_weight=unit_weight)
# check if same as sample_weight=None
assert_array_equal(clf_unitweight.predict(T), clf.predict(T))
assert_allclose(clf.coef_, clf_unitweight.coef_, 1, 0.0001)
# check that fit(X) = fit([X1, X2, X3],sample_weight = [n1, n2, n3]) where
# X = X1 repeated n1 times, X2 repeated n2 times and so forth
random_state = check_random_state(0)
random_weight = random_state.randint(0, 10, n_samples)
lsvc_unflat = svm.LinearSVC(random_state=0).\
fit(X, Y, sample_weight=random_weight)
pred1 = lsvc_unflat.predict(T)
X_flat = np.repeat(X, random_weight, axis=0)
y_flat = np.repeat(Y, random_weight, axis=0)
lsvc_flat = svm.LinearSVC(random_state=0).fit(X_flat, y_flat)
pred2 = lsvc_flat.predict(T)
assert_array_equal(pred1, pred2)
assert_allclose(lsvc_unflat.coef_, lsvc_flat.coef_, 1, 0.0001)
def test_linear_svx_uppercase_loss_penality_raises_error():
# Check if Upper case notation raises error at _fit_liblinear
# which is called by fit
X, y = [[0.0], [1.0]], [0, 1]
assert_raise_message(ValueError, "loss='SQuared_hinge' is not supported",
svm.LinearSVC(loss="SQuared_hinge").fit, X, y)
assert_raise_message(ValueError,
("The combination of penalty='L2'"
" and loss='squared_hinge' is not supported"),
svm.LinearSVC(penalty="L2").fit, X, y)
def test_auto_weight():
# Test class weights for imbalanced data
from mrex.linear_model import LogisticRegression
# We take as dataset the two-dimensional projection of iris so
# that it is not separable and remove half of predictors from
# class 1.
# We add one to the targets as a non-regression test:
# class_weight="balanced"
# used to work only when the labels where a range [0..K).
from mrex.utils import compute_class_weight
X, y = iris.data[:, :2], iris.target + 1
unbalanced = np.delete(np.arange(y.size), np.where(y > 2)[0][::2])
classes = np.unique(y[unbalanced])
class_weights = compute_class_weight('balanced', classes, y[unbalanced])
assert np.argmax(class_weights) == 2
for clf in (svm.SVC(kernel='linear'), svm.LinearSVC(random_state=0),
LogisticRegression()):
# check that score is better when class='balanced' is set.
y_pred = clf.fit(X[unbalanced], y[unbalanced]).predict(X)
clf.set_params(class_weight='balanced')
y_pred_balanced = clf.fit(
X[unbalanced],
y[unbalanced],
).predict(X)
assert (metrics.f1_score(y, y_pred, average='macro') <=
metrics.f1_score(y, y_pred_balanced, average='macro'))
def test_linearsvc():
# Similar to test_SVC
clf = svm.LinearSVC(random_state=0).fit(X, Y)
sp_clf = svm.LinearSVC(random_state=0).fit(X_sp, Y)
assert sp_clf.fit_intercept
assert_array_almost_equal(clf.coef_, sp_clf.coef_, decimal=4)
assert_array_almost_equal(clf.intercept_, sp_clf.intercept_, decimal=4)
assert_array_almost_equal(clf.predict(X), sp_clf.predict(X_sp))
clf.fit(X2, Y2)
sp_clf.fit(X2_sp, Y2)
assert_array_almost_equal(clf.coef_, sp_clf.coef_, decimal=4)
assert_array_almost_equal(clf.intercept_, sp_clf.intercept_, decimal=4)
def test_crammer_singer_binary():
# Test Crammer-Singer formulation in the binary case
X, y = make_classification(n_classes=2, random_state=0)
for fit_intercept in (True, False):
acc = svm.LinearSVC(fit_intercept=fit_intercept,
multi_class="crammer_singer",
random_state=0).fit(X, y).score(X, y)
assert acc > 0.9
def test_linearsvc_crammer_singer():
# Test LinearSVC with crammer_singer multi-class svm
ovr_clf = svm.LinearSVC(random_state=0).fit(iris.data, iris.target)
cs_clf = svm.LinearSVC(multi_class='crammer_singer', random_state=0)
cs_clf.fit(iris.data, iris.target)
# similar prediction for ovr and crammer-singer:
assert (ovr_clf.predict(iris.data) == cs_clf.predict(
iris.data)).mean() > .9
# classifiers shouldn't be the same
assert (ovr_clf.coef_ != cs_clf.coef_).all()
# test decision function
assert_array_equal(cs_clf.predict(iris.data),
np.argmax(cs_clf.decision_function(iris.data), axis=1))
dec_func = np.dot(iris.data, cs_clf.coef_.T) + cs_clf.intercept_
assert_array_almost_equal(dec_func, cs_clf.decision_function(iris.data))
def test_linear_svc_intercept_scaling():
# Test that the right error message is thrown when intercept_scaling <= 0
for i in [-1, 0]:
lsvc = svm.LinearSVC(intercept_scaling=i)
msg = ('Intercept scaling is %r but needs to be greater than 0.'
' To disable fitting an intercept,'
' set fit_intercept=False.' % lsvc.intercept_scaling)
assert_raise_message(ValueError, msg, lsvc.fit, X, Y)
def test_liblinear_set_coef():
# multi-class case
clf = svm.LinearSVC().fit(iris.data, iris.target)
values = clf.decision_function(iris.data)
clf.coef_ = clf.coef_.copy()
clf.intercept_ = clf.intercept_.copy()
values2 = clf.decision_function(iris.data)
assert_array_almost_equal(values, values2)
# binary-class case
X = [[2, 1], [3, 1], [1, 3], [2, 3]]
y = [0, 0, 1, 1]
clf = svm.LinearSVC().fit(X, y)
values = clf.decision_function(X)
clf.coef_ = clf.coef_.copy()
clf.intercept_ = clf.intercept_.copy()
values2 = clf.decision_function(X)
assert_array_equal(values, values2)
def test_linear_svm_convergence_warnings():
# Test that warnings are raised if model does not converge
lsvc = svm.LinearSVC(random_state=0, max_iter=2)
assert_warns(ConvergenceWarning, lsvc.fit, X, Y)
assert lsvc.n_iter_ == 2
lsvr = svm.LinearSVR(random_state=0, max_iter=2)
assert_warns(ConvergenceWarning, lsvr.fit, iris.data, iris.target)
assert lsvr.n_iter_ == 2
def test_linearsvc_iris():
# Test that LinearSVC gives plausible predictions on the iris dataset
# Also, test symbolic class names (classes_).
target = iris.target_names[iris.target]
clf = svm.LinearSVC(random_state=0).fit(iris.data, target)
assert set(clf.classes_) == set(iris.target_names)
assert np.mean(clf.predict(iris.data) == target) > 0.8
dec = clf.decision_function(iris.data)
pred = iris.target_names[np.argmax(dec, 1)]
assert_array_equal(pred, clf.predict(iris.data))
def test_linearsvc_verbose():
# stdout: redirect
import os
stdout = os.dup(1) # save original stdout
os.dup2(os.pipe()[1], 1) # replace it
# actual call
clf = svm.LinearSVC(verbose=1)
clf.fit(X, Y)
# stdout: restore
os.dup2(stdout, 1) # restore original stdout
def test_linearsvc_iris():
# Test the sparse LinearSVC with the iris dataset
sp_clf = svm.LinearSVC(random_state=0).fit(iris.data, iris.target)
clf = svm.LinearSVC(random_state=0).fit(iris.data.toarray(), iris.target)
assert clf.fit_intercept == sp_clf.fit_intercept
assert_array_almost_equal(clf.coef_, sp_clf.coef_, decimal=1)
assert_array_almost_equal(clf.intercept_, sp_clf.intercept_, decimal=1)
assert_array_almost_equal(clf.predict(iris.data.toarray()),
sp_clf.predict(iris.data))
# check decision_function
pred = np.argmax(sp_clf.decision_function(iris.data), 1)
assert_array_almost_equal(pred, clf.predict(iris.data.toarray()))
# sparsify the coefficients on both models and check that they still
# produce the same results
clf.sparsify()
assert_array_equal(pred, clf.predict(iris.data))
sp_clf.sparsify()
assert_array_equal(pred, sp_clf.predict(iris.data))
def test_weight():
# Test class weights
X_, y_ = make_classification(n_samples=200,
n_features=100,
weights=[0.833, 0.167],
random_state=0)
X_ = sparse.csr_matrix(X_)
for clf in (linear_model.LogisticRegression(),
svm.LinearSVC(random_state=0), svm.SVC()):
clf.set_params(class_weight={0: 5})
clf.fit(X_[:180], y_[:180])
y_pred = clf.predict(X_[180:])
assert np.sum(y_pred == y_[180:]) >= 11
def test_bad_input():
# Test that it gives proper exception on deficient input
# impossible value of C
with pytest.raises(ValueError):
svm.SVC(C=-1).fit(X, Y)
# impossible value of nu
clf = svm.NuSVC(nu=0.0)
with pytest.raises(ValueError):
clf.fit(X, Y)
Y2 = Y[:-1] # wrong dimensions for labels
with pytest.raises(ValueError):
clf.fit(X, Y2)
# Test with arrays that are non-contiguous.
for clf in (svm.SVC(), svm.LinearSVC(random_state=0)):
Xf = np.asfortranarray(X)
assert not Xf.flags['C_CONTIGUOUS']
yf = np.ascontiguousarray(np.tile(Y, (2, 1)).T)
yf = yf[:, -1]
assert not yf.flags['F_CONTIGUOUS']
assert not yf.flags['C_CONTIGUOUS']
clf.fit(Xf, yf)
assert_array_equal(clf.predict(T), true_result)
# error for precomputed kernelsx
clf = svm.SVC(kernel='precomputed')
with pytest.raises(ValueError):
clf.fit(X, Y)
# sample_weight bad dimensions
clf = svm.SVC()
with pytest.raises(ValueError):
clf.fit(X, Y, sample_weight=range(len(X) - 1))
# predict with sparse input when trained with dense
clf = svm.SVC().fit(X, Y)
with pytest.raises(ValueError):
clf.predict(sparse.lil_matrix(X))
Xt = np.array(X).T
clf.fit(np.dot(X, Xt), Y)
with pytest.raises(ValueError):
clf.predict(X)
clf = svm.SVC()
clf.fit(X, Y)
with pytest.raises(ValueError):
clf.predict(Xt)
def test_weight():
# Test class weights
clf = svm.SVC(class_weight={1: 0.1})
# we give a small weights to class 1
clf.fit(X, Y)
# so all predicted values belong to class 2
assert_array_almost_equal(clf.predict(X), [2] * 6)
X_, y_ = make_classification(n_samples=200,
n_features=10,
weights=[0.833, 0.167],
random_state=2)
for clf in (linear_model.LogisticRegression(),
svm.LinearSVC(random_state=0), svm.SVC()):
clf.set_params(class_weight={0: .1, 1: 10})
clf.fit(X_[:100], y_[:100])
y_pred = clf.predict(X_[100:])
assert f1_score(y_[100:], y_pred) > .3
n_samples = len(digits.data)
data = digits.data / 16.
data -= data.mean(axis=0)
# We learn the digits on the first half of the digits
data_train, targets_train = (data[:n_samples // 2],
digits.target[:n_samples // 2])
# Now predict the value of the digit on the second half:
data_test, targets_test = (data[n_samples // 2:],
digits.target[n_samples // 2:])
# data_test = scaler.transform(data_test)
# Create a classifier: a support vector classifier
kernel_svm = svm.SVC(gamma=.2)
linear_svm = svm.LinearSVC()
# create pipeline from kernel approximation
# and linear svm
feature_map_fourier = RBFSampler(gamma=.2, random_state=1)
feature_map_nystroem = Nystroem(gamma=.2, random_state=1)
fourier_approx_svm = pipeline.Pipeline([("feature_map", feature_map_fourier),
("svm", svm.LinearSVC())])
nystroem_approx_svm = pipeline.Pipeline([("feature_map", feature_map_nystroem),
("svm", svm.LinearSVC())])
# fit and predict using linear and kernel svm:
kernel_svm_time = time()
kernel_svm.fit(data_train, targets_train)
def test_lsvc_intercept_scaling_zero():
# Test that intercept_scaling is ignored when fit_intercept is False
lsvc = svm.LinearSVC(fit_intercept=False)
lsvc.fit(X, Y)
assert lsvc.intercept_ == 0.
from mrex.datasets import samples_generator
from mrex.feature_selection import SelectKBest, f_regression
from mrex.pipeline import make_pipeline
from mrex.model_selection import train_test_split
from mrex.metrics import classification_report
print(__doc__)
# import some data to play with
X, y = samples_generator.make_classification(n_features=20,
n_informative=3,
n_redundant=0,
n_classes=4,
n_clusters_per_class=2)
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42)
# ANOVA SVM-C
# 1) anova filter, take 3 best ranked features
anova_filter = SelectKBest(f_regression, k=3)
# 2) svm
clf = svm.LinearSVC()
anova_svm = make_pipeline(anova_filter, clf)
anova_svm.fit(X_train, y_train)
y_pred = anova_svm.predict(X_test)
print(classification_report(y_test, y_pred))
coef = anova_svm[:-1].inverse_transform(anova_svm['linearsvc'].coef_)
print(coef)
X = iris.data
y = iris.target
# Add noisy features
random_state = np.random.RandomState(0)
n_samples, n_features = X.shape
X = np.c_[X, random_state.randn(n_samples, 200 * n_features)]
# Limit to the two first classes, and split into training and test
X_train, X_test, y_train, y_test = train_test_split(X[y < 2],
y[y < 2],
test_size=.5,
random_state=random_state)
# Create a simple classifier
classifier = svm.LinearSVC(random_state=random_state)
classifier.fit(X_train, y_train)
y_score = classifier.decision_function(X_test)
###############################################################################
# Compute the average precision score
# ...................................
from mrex.metrics import average_precision_score
average_precision = average_precision_score(y_test, y_score)
print('Average precision-recall score: {0:0.2f}'.format(average_precision))
###############################################################################
# Plot the Precision-Recall curve
# ................................
from mrex.metrics import precision_recall_curve
Z = clf.predict(np.c_[xx.ravel(), yy.ravel()])
Z = Z.reshape(xx.shape)
out = ax.contourf(xx, yy, Z, **params)
return out
# import some data to play with
iris = datasets.load_iris()
# Take the first two features. We could avoid this by using a two-dim dataset
X = iris.data[:, :2]
y = iris.target
# we create an instance of SVM and fit out data. We do not scale our
# data since we want to plot the support vectors
C = 1.0 # SVM regularization parameter
models = (svm.SVC(kernel='linear', C=C), svm.LinearSVC(C=C, max_iter=10000),
svm.SVC(kernel='rbf', gamma=0.7,
C=C), svm.SVC(kernel='poly', degree=3, gamma='auto', C=C))
models = (clf.fit(X, y) for clf in models)
# title for the plots
titles = ('SVC with linear kernel', 'LinearSVC (linear kernel)',
'SVC with RBF kernel', 'SVC with polynomial (degree 3) kernel')
# Set-up 2x2 grid for plotting.
fig, sub = plt.subplots(2, 2)
plt.subplots_adjust(wspace=0.4, hspace=0.4)
X0, X1 = X[:, 0], X[:, 1]
xx, yy = make_meshgrid(X0, X1)