def test_dense_liblinear_intercept_handling(classifier=svm.LinearSVC):
    # Test that dense liblinear honours intercept_scaling param
    X = [[2, 1],
         [3, 1],
         [1, 3],
         [2, 3]]
    y = [0, 0, 1, 1]
    clf = classifier(fit_intercept=True, penalty='l1', loss='squared_hinge',
                     dual=False, C=4, tol=1e-7, random_state=0)
    assert_true(clf.intercept_scaling == 1, clf.intercept_scaling)
    assert_true(clf.fit_intercept)

    # when intercept_scaling is low the intercept value is highly "penalized"
    # by regularization
    clf.intercept_scaling = 1
    clf.fit(X, y)
    assert_almost_equal(clf.intercept_, 0, decimal=5)

    # when intercept_scaling is sufficiently high, the intercept value
    # is not affected by regularization
    clf.intercept_scaling = 100
    clf.fit(X, y)
    intercept1 = clf.intercept_
    assert_less(intercept1, -1)

    # when intercept_scaling is sufficiently high, the intercept value
    # doesn't depend on intercept_scaling value
    clf.intercept_scaling = 1000
    clf.fit(X, y)
    intercept2 = clf.intercept_
    assert_array_almost_equal(intercept1, intercept2, decimal=2)

def test_labels_assignment_and_inertia():
    # pure numpy implementation as easily auditable reference gold
    # implementation
    rng = np.random.RandomState(42)
    noisy_centers = centers + rng.normal(size=centers.shape)
    labels_gold = - np.ones(n_samples, dtype=np.int)
    mindist = np.empty(n_samples)
    mindist.fill(np.infty)
    for center_id in range(n_clusters):
        dist = np.sum((X - noisy_centers[center_id]) ** 2, axis=1)
        labels_gold[dist < mindist] = center_id
        mindist = np.minimum(dist, mindist)
    inertia_gold = mindist.sum()
    assert_true((mindist >= 0.0).all())
    assert_true((labels_gold != -1).all())

    # perform label assignment using the dense array input
    x_squared_norms = (X ** 2).sum(axis=1)
    labels_array, inertia_array = _labels_inertia(
        X, x_squared_norms, noisy_centers)
    assert_array_almost_equal(inertia_array, inertia_gold)
    assert_array_equal(labels_array, labels_gold)

    # perform label assignment using the sparse CSR input
    x_squared_norms_from_csr = row_norms(X_csr, squared=True)
    labels_csr, inertia_csr = _labels_inertia(
        X_csr, x_squared_norms_from_csr, noisy_centers)
    assert_array_almost_equal(inertia_csr, inertia_gold)
    assert_array_equal(labels_csr, labels_gold)

def test_grid_search_precomputed_kernel():
    """Test that grid search works when the input features are given in the
    form of a precomputed kernel matrix """
    X_, y_ = make_classification(n_samples=200, n_features=100, random_state=0)

    # compute the training kernel matrix corresponding to the linear kernel
    K_train = np.dot(X_[:180], X_[:180].T)
    y_train = y_[:180]

    clf = SVC(kernel='precomputed')
    cv = GridSearchCV(clf, {'C': [0.1, 1.0]})
    cv.fit(K_train, y_train)

    assert_true(cv.best_score_ >= 0)

    # compute the test kernel matrix
    K_test = np.dot(X_[180:], X_[:180].T)
    y_test = y_[180:]

    y_pred = cv.predict(K_test)

    assert_true(np.mean(y_pred == y_test) >= 0)

    # test error is raised when the precomputed kernel is not array-like
    # or sparse
    assert_raises(ValueError, cv.fit, K_train.tolist(), y_train)

def test_ovr_multilabel():
    # Toy dataset where features correspond directly to labels.
    X = np.array([[0, 4, 5], [0, 5, 0], [3, 3, 3], [4, 0, 6], [6, 0, 0]])
    y = [["spam", "eggs"], ["spam"], ["ham", "eggs", "spam"],
         ["ham", "eggs"], ["ham"]]
    #y = [[1, 2], [1], [0, 1, 2], [0, 2], [0]]
    Y = np.array([[0, 1, 1],
                  [0, 1, 0],
                  [1, 1, 1],
                  [1, 0, 1],
                  [1, 0, 0]])

    classes = set("ham eggs spam".split())

    for base_clf in (MultinomialNB(), LinearSVC(random_state=0),
                     LinearRegression(), Ridge(),
                     ElasticNet(), Lasso(alpha=0.5)):
        # test input as lists of tuples
        clf = assert_warns(DeprecationWarning,
                           OneVsRestClassifier(base_clf).fit,
                           X, y)
        assert_equal(set(clf.classes_), classes)
        y_pred = clf.predict([[0, 4, 4]])[0]
        assert_equal(set(y_pred), set(["spam", "eggs"]))
        assert_true(clf.multilabel_)

        # test input as label indicator matrix
        clf = OneVsRestClassifier(base_clf).fit(X, Y)
        y_pred = clf.predict([[0, 4, 4]])[0]
        assert_array_equal(y_pred, [0, 1, 1])
        assert_true(clf.multilabel_)

def train_test_split_mock_pandas():
    # X mock dataframe
    X_df = MockDataFrame(X)
    X_train, X_test = train_test_split(X_df)
    assert_true(isinstance(X_train, MockDataFrame))
    assert_true(isinstance(X_test, MockDataFrame))
    X_train_arr, X_test_arr = train_test_split(X_df)

def test_symmetry():
    """Test the symmetry of score and loss functions"""
    random_state = check_random_state(0)
    y_true = random_state.randint(0, 2, size=(20, ))
    y_pred = random_state.randint(0, 2, size=(20, ))

    # We shouldn't forget any metrics
    assert_equal(set(SYMMETRIC_METRICS).union(NOT_SYMMETRIC_METRICS,
                                              THRESHOLDED_METRICS,
                                              METRIC_UNDEFINED_MULTICLASS),
                 set(ALL_METRICS))

    assert_equal(
        set(SYMMETRIC_METRICS).intersection(set(NOT_SYMMETRIC_METRICS)),
        set([]))

    # Symmetric metric
    for name in SYMMETRIC_METRICS:
        metric = ALL_METRICS[name]
        assert_almost_equal(metric(y_true, y_pred),
                            metric(y_pred, y_true),
                            err_msg="%s is not symmetric" % name)

    # Not symmetric metrics
    for name in NOT_SYMMETRIC_METRICS:
        metric = ALL_METRICS[name]
        assert_true(np.any(metric(y_true, y_pred) != metric(y_pred, y_true)),
                    msg="%s seems to be symmetric" % name)

def check_get_params_invariance(name, estimator):
    class T(BaseEstimator):
        """Mock classifier
        """

        def __init__(self):
            pass

        def fit(self, X, y):
            return self

    if name in ('FeatureUnion', 'Pipeline'):
        e = estimator([('clf', T())])

    elif name in ('GridSearchCV' 'RandomizedSearchCV'):
        return

    else:
        e = estimator()

    shallow_params = e.get_params(deep=False)
    deep_params = e.get_params(deep=True)

    assert_true(all(item in deep_params.items() for item in
                    shallow_params.items()))

def test_enet_path_positive():
    # Test that the coefs returned by positive=True in enet_path are positive

    X, y, _, _ = build_dataset(n_samples=50, n_features=50)
    for path in [enet_path, lasso_path]:
        pos_path_coef = path(X, y, positive=True)[1]
        assert_true(np.all(pos_path_coef >= 0))

    def test_sgd_l1(self):
        """Test L1 regularization"""
        n = len(X4)
        rng = np.random.RandomState(13)
        idx = np.arange(n)
        rng.shuffle(idx)

        X = X4[idx, :]
        Y = Y4[idx]

        clf = self.factory(penalty="l1", alpha=0.2, fit_intercept=False, n_iter=2000, shuffle=False)
        clf.fit(X, Y)
        assert_array_equal(clf.coef_[0, 1:-1], np.zeros((4,)))
        pred = clf.predict(X)
        assert_array_equal(pred, Y)

        # test sparsify with dense inputs
        clf.sparsify()
        assert_true(sp.issparse(clf.coef_))
        pred = clf.predict(X)
        assert_array_equal(pred, Y)

        # pickle and unpickle with sparse coef_
        clf = pickle.loads(pickle.dumps(clf))
        assert_true(sp.issparse(clf.coef_))
        pred = clf.predict(X)
        assert_array_equal(pred, Y)

def test_check_increasing_up_extreme():
    x = [0, 1, 2, 3, 4, 5]
    y = [0, 1, 2, 3, 4, 5]

    # Check that we got increasing=True and no warnings
    is_increasing = assert_no_warnings(check_increasing, x, y)
    assert_true(is_increasing)

def test_lasso_cv():
    X, y, X_test, y_test = build_dataset()
    max_iter = 150
    clf = LassoCV(n_alphas=10, eps=1e-3, max_iter=max_iter).fit(X, y)
    assert_almost_equal(clf.alpha_, 0.056, 2)

    clf = LassoCV(n_alphas=10, eps=1e-3, max_iter=max_iter, precompute=True)
    clf.fit(X, y)
    assert_almost_equal(clf.alpha_, 0.056, 2)

    # Check that the lars and the coordinate descent implementation
    # select a similar alpha
    lars = LassoLarsCV(normalize=False, max_iter=30).fit(X, y)
    # for this we check that they don't fall in the grid of
    # clf.alphas further than 1
    assert_true(np.abs(
        np.searchsorted(clf.alphas_[::-1], lars.alpha_) -
        np.searchsorted(clf.alphas_[::-1], clf.alpha_)) <= 1)
    # check that they also give a similar MSE
    mse_lars = interpolate.interp1d(lars.cv_alphas_, lars.cv_mse_path_.T)
    np.testing.assert_approx_equal(mse_lars(clf.alphas_[5]).mean(),
                                   clf.mse_path_[5].mean(), significant=2)

    # test set
    assert_greater(clf.score(X_test, y_test), 0.99)

def test_check_increasing_up():
    x = [0, 1, 2, 3, 4, 5]
    y = [0, 1.5, 2.77, 8.99, 8.99, 50]

    # Check that we got increasing=True and no warnings
    is_increasing = assert_no_warnings(check_increasing, x, y)
    assert_true(is_increasing)

def test_fit_transform():
    alpha = 1
    rng = np.random.RandomState(0)
    Y, _, _ = generate_toy_data(3, 10, (8, 8), random_state=rng)  # wide array
    spca_lars = SparsePCA(n_components=3, method='lars', alpha=alpha,
                          random_state=0)
    spca_lars.fit(Y)
    U1 = spca_lars.transform(Y)
    # Test multiple CPUs
    if sys.platform == 'win32':  # fake parallelism for win32
        import sklearn.externals.joblib.parallel as joblib_par
        _mp = joblib_par.multiprocessing
        joblib_par.multiprocessing = None
        try:
            spca = SparsePCA(n_components=3, n_jobs=2, random_state=0,
                             alpha=alpha).fit(Y)
            U2 = spca.transform(Y)
        finally:
            joblib_par.multiprocessing = _mp
    else:  # we can efficiently use parallelism
        spca = SparsePCA(n_components=3, n_jobs=2, method='lars', alpha=alpha,
                         random_state=0).fit(Y)
        U2 = spca.transform(Y)
    assert_true(not np.all(spca_lars.components_ == 0))
    assert_array_almost_equal(U1, U2)
    # Test that CD gives similar results
    spca_lasso = SparsePCA(n_components=3, method='cd', random_state=0,
                           alpha=alpha)
    spca_lasso.fit(Y)
    assert_array_almost_equal(spca_lasso.components_, spca_lars.components_)

def test_scale_function_without_centering():
    rng = np.random.RandomState(42)
    X = rng.randn(4, 5)
    X[:, 0] = 0.0  # first feature is always of zero
    X_csr = sparse.csr_matrix(X)

    X_scaled = scale(X, with_mean=False)
    assert_false(np.any(np.isnan(X_scaled)))

    X_csr_scaled = scale(X_csr, with_mean=False)
    assert_false(np.any(np.isnan(X_csr_scaled.data)))

    # test csc has same outcome
    X_csc_scaled = scale(X_csr.tocsc(), with_mean=False)
    assert_array_almost_equal(X_scaled, X_csc_scaled.toarray())

    # raises value error on axis != 0
    assert_raises(ValueError, scale, X_csr, with_mean=False, axis=1)

    assert_array_almost_equal(X_scaled.mean(axis=0),
                              [0., -0.01, 2.24, -0.35, -0.78], 2)
    assert_array_almost_equal(X_scaled.std(axis=0), [0., 1., 1., 1., 1.])
    # Check that X has not been copied
    assert_true(X_scaled is not X)

    X_csr_scaled_mean, X_csr_scaled_std = mean_variance_axis0(X_csr_scaled)
    assert_array_almost_equal(X_csr_scaled_mean, X_scaled.mean(axis=0))
    assert_array_almost_equal(X_csr_scaled_std, X_scaled.std(axis=0))

def test_neighbors_accuracy_with_n_estimators():
    # Checks whether accuracy increases as `n_estimators` increases.
    n_estimators = np.array([1, 10, 100])
    n_samples = 100
    n_features = 10
    n_iter = 10
    n_points = 5
    rng = np.random.RandomState(42)
    accuracies = np.zeros(n_estimators.shape[0], dtype=float)
    X = rng.rand(n_samples, n_features)

    for i, t in enumerate(n_estimators):
        lshf = ignore_warnings(LSHForest, category=DeprecationWarning)(
            n_candidates=500, n_estimators=t)
        ignore_warnings(lshf.fit)(X)
        for j in range(n_iter):
            query = X[rng.randint(0, n_samples)].reshape(1, -1)
            neighbors = lshf.kneighbors(query, n_neighbors=n_points,
                                        return_distance=False)
            distances = pairwise_distances(query, X, metric='cosine')
            ranks = np.argsort(distances)[0, :n_points]

            intersection = np.intersect1d(ranks, neighbors).shape[0]
            ratio = intersection / float(n_points)
            accuracies[i] = accuracies[i] + ratio

        accuracies[i] = accuracies[i] / float(n_iter)
    # Sorted accuracies should be equal to original accuracies
    assert_true(np.all(np.diff(accuracies) >= 0),
                msg="Accuracies are not non-decreasing.")
    # Highest accuracy should be strictly greater than the lowest
    assert_true(np.ptp(accuracies) > 0,
                msg="Highest accuracy is not strictly greater than lowest.")

def test_simple():
    # Principle of Lars is to keep covariances tied and decreasing

    # also test verbose output
    from sklearn.externals.six.moves import cStringIO as StringIO
    import sys
    old_stdout = sys.stdout
    try:
        sys.stdout = StringIO()

        alphas_, active, coef_path_ = linear_model.lars_path(
            diabetes.data, diabetes.target, method="lar", verbose=10)

        sys.stdout = old_stdout

        for (i, coef_) in enumerate(coef_path_.T):
            res = y - np.dot(X, coef_)
            cov = np.dot(X.T, res)
            C = np.max(abs(cov))
            eps = 1e-3
            ocur = len(cov[C - eps < abs(cov)])
            if i < X.shape[1]:
                assert_true(ocur == i + 1)
            else:
                # no more than max_pred variables can go into the active set
                assert_true(ocur == X.shape[1])
    finally:
        sys.stdout = old_stdout

def test_kfold_valueerrors():
    # Check that errors are raised if there is not enough samples
    assert_raises(ValueError, cval.KFold, 3, 4)

    # Check that a warning is raised if the least populated class has too few
    # members.
    with warnings.catch_warnings(record=True) as w:
        warnings.simplefilter('always')
        y = [3, 3, -1, -1, 2]
        cv = cval.StratifiedKFold(y, 3)
        # checking there was only one warning.
        assert_equal(len(w), 1)
        # checking it has the right type
        assert_equal(w[0].category, Warning)
        # checking it's the right warning. This might be a bad test since it's
        # a characteristic of the code and not a behavior
        assert_true("The least populated class" in str(w[0]))

        # Check that despite the warning the folds are still computed even
        # though all the classes are not necessarily represented at on each
        # side of the split at each split
        check_cv_coverage(cv, expected_n_iter=3, n_samples=len(y))

    # Error when number of folds is <= 1
    assert_raises(ValueError, cval.KFold, 2, 0)
    assert_raises(ValueError, cval.KFold, 2, 1)
    assert_raises(ValueError, cval.StratifiedKFold, y, 0)
    assert_raises(ValueError, cval.StratifiedKFold, y, 1)

    # When n is not integer:
    assert_raises(ValueError, cval.KFold, 2.5, 2)

    # When n_folds is not integer:
    assert_raises(ValueError, cval.KFold, 5, 1.5)
    assert_raises(ValueError, cval.StratifiedKFold, y, 1.5)

def test_lars_path_positive_constraint():
    # this is the main test for the positive parameter on the lars_path method
    # the estimator classes just make use of this function

    # we do the test on the diabetes dataset

    # ensure that we get negative coefficients when positive=False
    # and all positive when positive=True
    # for method 'lar' (default) and lasso

    # Once deprecation of LAR + positive option is done use these:
    # assert_raises(ValueError, linear_model.lars_path, diabetes['data'],
    #               diabetes['target'], method='lar', positive=True)

    with pytest.warns(DeprecationWarning, match="broken"):
        linear_model.lars_path(diabetes['data'], diabetes['target'],
                               return_path=True, method='lar',
                               positive=True)

    method = 'lasso'
    alpha, active, coefs = \
        linear_model.lars_path(diabetes['data'], diabetes['target'],
                               return_path=True, method=method,
                               positive=False)
    assert_true(coefs.min() < 0)

    alpha, active, coefs = \
        linear_model.lars_path(diabetes['data'], diabetes['target'],
                               return_path=True, method=method,
                               positive=True)
    assert_true(coefs.min() >= 0)

def check_warm_start_oob(name):
    # Test that the warm start computes oob score when asked.
    X, y = hastie_X, hastie_y
    ForestEstimator = FOREST_ESTIMATORS[name]
    # Use 15 estimators to avoid 'some inputs do not have OOB scores' warning.
    clf = ForestEstimator(n_estimators=15, max_depth=3, warm_start=False,
                          random_state=1, bootstrap=True, oob_score=True)
    clf.fit(X, y)

    clf_2 = ForestEstimator(n_estimators=5, max_depth=3, warm_start=False,
                            random_state=1, bootstrap=True, oob_score=False)
    clf_2.fit(X, y)

    clf_2.set_params(warm_start=True, oob_score=True, n_estimators=15)
    clf_2.fit(X, y)

    assert_true(hasattr(clf_2, 'oob_score_'))
    assert_equal(clf.oob_score_, clf_2.oob_score_)

    # Test that oob_score is computed even if we don't need to train
    # additional trees.
    clf_3 = ForestEstimator(n_estimators=15, max_depth=3, warm_start=True,
                            random_state=1, bootstrap=True, oob_score=False)
    clf_3.fit(X, y)
    assert_true(not(hasattr(clf_3, 'oob_score_')))

    clf_3.set_params(oob_score=True)
    ignore_warnings(clf_3.fit)(X, y)

    assert_equal(clf.oob_score_, clf_3.oob_score_)

def test_grid_search_score_method():
    X, y = make_classification(n_samples=100, n_classes=2, flip_y=.2,
                               random_state=0)
    clf = LinearSVC(random_state=0)
    grid = {'C': [.1]}

    search_no_scoring = GridSearchCV(clf, grid, scoring=None).fit(X, y)
    search_accuracy = GridSearchCV(clf, grid, scoring='accuracy').fit(X, y)
    search_no_score_method_auc = GridSearchCV(LinearSVCNoScore(), grid,
                                              scoring='roc_auc').fit(X, y)
    search_auc = GridSearchCV(clf, grid, scoring='roc_auc').fit(X, y)

    # ChangedBehaviourWarning occurred previously (prior to #9005)
    score_no_scoring = assert_no_warnings(search_no_scoring.score, X, y)
    score_accuracy = assert_no_warnings(search_accuracy.score, X, y)
    score_no_score_auc = assert_no_warnings(search_no_score_method_auc.score,
                                            X, y)
    score_auc = assert_no_warnings(search_auc.score, X, y)

    # ensure the test is sane
    assert_true(score_auc < 1.0)
    assert_true(score_accuracy < 1.0)
    assert_not_equal(score_auc, score_accuracy)

    assert_almost_equal(score_accuracy, score_no_scoring)
    assert_almost_equal(score_auc, score_no_score_auc)