test_multioutput.py

import pytest
import numpy as np
import scipy.sparse as sp
from joblib import cpu_count

from sklearn.utils.testing import assert_almost_equal
from sklearn.utils.testing import assert_raises
from sklearn.utils.testing import assert_raises_regex
from sklearn.utils.testing import assert_raise_message
from sklearn.utils.testing import assert_array_equal
from sklearn.utils.testing import assert_array_almost_equal
from sklearn import datasets
from sklearn.base import clone
from sklearn.datasets import make_classification
from sklearn.ensemble import GradientBoostingRegressor, RandomForestClassifier
from sklearn.exceptions import NotFittedError
from sklearn.linear_model import Lasso
from sklearn.linear_model import LogisticRegression
from sklearn.linear_model import Ridge
from sklearn.linear_model import SGDClassifier
from sklearn.linear_model import SGDRegressor
from sklearn.metrics import jaccard_score, mean_squared_error
from sklearn.multiclass import OneVsRestClassifier
from sklearn.multioutput import ClassifierChain, RegressorChain
from sklearn.multioutput import MultiOutputClassifier
from sklearn.multioutput import MultiOutputRegressor
from sklearn.svm import LinearSVC
from sklearn.base import ClassifierMixin
from sklearn.utils import shuffle
from sklearn.model_selection import GridSearchCV


def test_multi_target_regression():
    X, y = datasets.make_regression(n_targets=3)
    X_train, y_train = X[:50], y[:50]
    X_test, y_test = X[50:], y[50:]

    references = np.zeros_like(y_test)
    for n in range(3):
        rgr = GradientBoostingRegressor(random_state=0)
        rgr.fit(X_train, y_train[:, n])
        references[:, n] = rgr.predict(X_test)

    rgr = MultiOutputRegressor(GradientBoostingRegressor(random_state=0))
    rgr.fit(X_train, y_train)
    y_pred = rgr.predict(X_test)

    assert_almost_equal(references, y_pred)


# 0.23. warning about tol not having its correct default value.
@pytest.mark.filterwarnings('ignore:max_iter and tol parameters have been')
def test_multi_target_regression_partial_fit():
    X, y = datasets.make_regression(n_targets=3)
    X_train, y_train = X[:50], y[:50]
    X_test, y_test = X[50:], y[50:]

    references = np.zeros_like(y_test)
    half_index = 25
    for n in range(3):
        sgr = SGDRegressor(random_state=0, max_iter=5)
        sgr.partial_fit(X_train[:half_index], y_train[:half_index, n])
        sgr.partial_fit(X_train[half_index:], y_train[half_index:, n])
        references[:, n] = sgr.predict(X_test)

    sgr = MultiOutputRegressor(SGDRegressor(random_state=0, max_iter=5))

    sgr.partial_fit(X_train[:half_index], y_train[:half_index])
    sgr.partial_fit(X_train[half_index:], y_train[half_index:])

    y_pred = sgr.predict(X_test)
    assert_almost_equal(references, y_pred)
    assert not hasattr(MultiOutputRegressor(Lasso), 'partial_fit')


def test_multi_target_regression_one_target():
    # Test multi target regression raises
    X, y = datasets.make_regression(n_targets=1)
    rgr = MultiOutputRegressor(GradientBoostingRegressor(random_state=0))
    assert_raises(ValueError, rgr.fit, X, y)


def test_multi_target_sparse_regression():
    X, y = datasets.make_regression(n_targets=3)
    X_train, y_train = X[:50], y[:50]
    X_test = X[50:]

    for sparse in [sp.csr_matrix, sp.csc_matrix, sp.coo_matrix, sp.dok_matrix,
                   sp.lil_matrix]:
        rgr = MultiOutputRegressor(Lasso(random_state=0))
        rgr_sparse = MultiOutputRegressor(Lasso(random_state=0))

        rgr.fit(X_train, y_train)
        rgr_sparse.fit(sparse(X_train), y_train)

        assert_almost_equal(rgr.predict(X_test),
                            rgr_sparse.predict(sparse(X_test)))


def test_multi_target_sample_weights_api():
    X = [[1, 2, 3], [4, 5, 6]]
    y = [[3.141, 2.718], [2.718, 3.141]]
    w = [0.8, 0.6]

    rgr = MultiOutputRegressor(Lasso())
    assert_raises_regex(ValueError, "does not support sample weights",
                        rgr.fit, X, y, w)

    # no exception should be raised if the base estimator supports weights
    rgr = MultiOutputRegressor(GradientBoostingRegressor(random_state=0))
    rgr.fit(X, y, w)


# 0.23. warning about tol not having its correct default value.
@pytest.mark.filterwarnings('ignore:max_iter and tol parameters have been')
def test_multi_target_sample_weight_partial_fit():
    # weighted regressor
    X = [[1, 2, 3], [4, 5, 6]]
    y = [[3.141, 2.718], [2.718, 3.141]]
    w = [2., 1.]
    rgr_w = MultiOutputRegressor(SGDRegressor(random_state=0, max_iter=5))
    rgr_w.partial_fit(X, y, w)

    # weighted with different weights
    w = [2., 2.]
    rgr = MultiOutputRegressor(SGDRegressor(random_state=0, max_iter=5))
    rgr.partial_fit(X, y, w)

    assert rgr.predict(X)[0][0] != rgr_w.predict(X)[0][0]


def test_multi_target_sample_weights():
    # weighted regressor
    Xw = [[1, 2, 3], [4, 5, 6]]
    yw = [[3.141, 2.718], [2.718, 3.141]]
    w = [2., 1.]
    rgr_w = MultiOutputRegressor(GradientBoostingRegressor(random_state=0))
    rgr_w.fit(Xw, yw, w)

    # unweighted, but with repeated samples
    X = [[1, 2, 3], [1, 2, 3], [4, 5, 6]]
    y = [[3.141, 2.718], [3.141, 2.718], [2.718, 3.141]]
    rgr = MultiOutputRegressor(GradientBoostingRegressor(random_state=0))
    rgr.fit(X, y)

    X_test = [[1.5, 2.5, 3.5], [3.5, 4.5, 5.5]]
    assert_almost_equal(rgr.predict(X_test), rgr_w.predict(X_test))


# Import the data
iris = datasets.load_iris()
# create a multiple targets by randomized shuffling and concatenating y.
X = iris.data
y1 = iris.target
y2 = shuffle(y1, random_state=1)
y3 = shuffle(y1, random_state=2)
y = np.column_stack((y1, y2, y3))
n_samples, n_features = X.shape
n_outputs = y.shape[1]
n_classes = len(np.unique(y1))
classes = list(map(np.unique, (y1, y2, y3)))


def test_multi_output_classification_partial_fit_parallelism():
    sgd_linear_clf = SGDClassifier(loss='log', random_state=1, max_iter=5)
    mor = MultiOutputClassifier(sgd_linear_clf, n_jobs=4)
    mor.partial_fit(X, y, classes)
    est1 = mor.estimators_[0]
    mor.partial_fit(X, y)
    est2 = mor.estimators_[0]
    if cpu_count() > 1:
        # parallelism requires this to be the case for a sane implementation
        assert est1 is not est2


# check predict_proba passes
def test_multi_output_predict_proba():
    sgd_linear_clf = SGDClassifier(random_state=1, max_iter=5)
    param = {'loss': ('hinge', 'log', 'modified_huber')}

    # inner function for custom scoring
    def custom_scorer(estimator, X, y):
        if hasattr(estimator, "predict_proba"):
            return 1.0
        else:
            return 0.0
    grid_clf = GridSearchCV(sgd_linear_clf, param_grid=param,
                            scoring=custom_scorer, cv=3)
    multi_target_linear = MultiOutputClassifier(grid_clf)
    multi_target_linear.fit(X, y)

    multi_target_linear.predict_proba(X)

    # SGDClassifier defaults to loss='hinge' which is not a probabilistic
    # loss function; therefore it does not expose a predict_proba method
    sgd_linear_clf = SGDClassifier(random_state=1, max_iter=5)
    multi_target_linear = MultiOutputClassifier(sgd_linear_clf)
    multi_target_linear.fit(X, y)
    err_msg = "The base estimator should implement predict_proba method"
    with pytest.raises(ValueError, match=err_msg):
        multi_target_linear.predict_proba(X)


# 0.23. warning about tol not having its correct default value.
@pytest.mark.filterwarnings('ignore:max_iter and tol parameters have been')
def test_multi_output_classification_partial_fit():
    # test if multi_target initializes correctly with base estimator and fit
    # assert predictions work as expected for predict

    sgd_linear_clf = SGDClassifier(loss='log', random_state=1, max_iter=5)
    multi_target_linear = MultiOutputClassifier(sgd_linear_clf)

    # train the multi_target_linear and also get the predictions.
    half_index = X.shape[0] // 2
    multi_target_linear.partial_fit(
        X[:half_index], y[:half_index], classes=classes)

    first_predictions = multi_target_linear.predict(X)
    assert (n_samples, n_outputs) == first_predictions.shape

    multi_target_linear.partial_fit(X[half_index:], y[half_index:])
    second_predictions = multi_target_linear.predict(X)
    assert (n_samples, n_outputs) == second_predictions.shape

    # train the linear classification with each column and assert that
    # predictions are equal after first partial_fit and second partial_fit
    for i in range(3):
        # create a clone with the same state
        sgd_linear_clf = clone(sgd_linear_clf)
        sgd_linear_clf.partial_fit(
            X[:half_index], y[:half_index, i], classes=classes[i])
        assert_array_equal(sgd_linear_clf.predict(X), first_predictions[:, i])
        sgd_linear_clf.partial_fit(X[half_index:], y[half_index:, i])
        assert_array_equal(sgd_linear_clf.predict(X), second_predictions[:, i])


# 0.23. warning about tol not having its correct default value.
@pytest.mark.filterwarnings('ignore:max_iter and tol parameters have been')
def test_multi_output_classification_partial_fit_no_first_classes_exception():
    sgd_linear_clf = SGDClassifier(loss='log', random_state=1, max_iter=5)
    multi_target_linear = MultiOutputClassifier(sgd_linear_clf)
    assert_raises_regex(ValueError, "classes must be passed on the first call "
                                    "to partial_fit.",
                        multi_target_linear.partial_fit, X, y)


def test_multi_output_classification():
    # test if multi_target initializes correctly with base estimator and fit
    # assert predictions work as expected for predict, prodict_proba and score

    forest = RandomForestClassifier(n_estimators=10, random_state=1)
    multi_target_forest = MultiOutputClassifier(forest)

    # train the multi_target_forest and also get the predictions.
    multi_target_forest.fit(X, y)

    predictions = multi_target_forest.predict(X)
    assert (n_samples, n_outputs) == predictions.shape

    predict_proba = multi_target_forest.predict_proba(X)

    assert len(predict_proba) == n_outputs
    for class_probabilities in predict_proba:
        assert (n_samples, n_classes) == class_probabilities.shape

    assert_array_equal(np.argmax(np.dstack(predict_proba), axis=1),
                       predictions)

    # train the forest with each column and assert that predictions are equal
    for i in range(3):
        forest_ = clone(forest)  # create a clone with the same state
        forest_.fit(X, y[:, i])
        assert list(forest_.predict(X)) == list(predictions[:, i])
        assert_array_equal(list(forest_.predict_proba(X)),
                           list(predict_proba[i]))


def test_multiclass_multioutput_estimator():
    # test to check meta of meta estimators
    svc = LinearSVC(random_state=0)
    multi_class_svc = OneVsRestClassifier(svc)
    multi_target_svc = MultiOutputClassifier(multi_class_svc)

    multi_target_svc.fit(X, y)

    predictions = multi_target_svc.predict(X)
    assert (n_samples, n_outputs) == predictions.shape

    # train the forest with each column and assert that predictions are equal
    for i in range(3):
        multi_class_svc_ = clone(multi_class_svc)  # create a clone
        multi_class_svc_.fit(X, y[:, i])
        assert (list(multi_class_svc_.predict(X)) ==
                     list(predictions[:, i]))


def test_multiclass_multioutput_estimator_predict_proba():
    seed = 542

    # make test deterministic
    rng = np.random.RandomState(seed)

    # random features
    X = rng.normal(size=(5, 5))

    # random labels
    y1 = np.array(['b', 'a', 'a', 'b', 'a']).reshape(5, 1)  # 2 classes
    y2 = np.array(['d', 'e', 'f', 'e', 'd']).reshape(5, 1)  # 3 classes

    Y = np.concatenate([y1, y2], axis=1)

    clf = MultiOutputClassifier(LogisticRegression(
        solver='liblinear', random_state=seed))

    clf.fit(X, Y)

    y_result = clf.predict_proba(X)
    y_actual = [np.array([[0.23481764, 0.76518236],
                          [0.67196072, 0.32803928],
                          [0.54681448, 0.45318552],
                          [0.34883923, 0.65116077],
                          [0.73687069, 0.26312931]]),
                np.array([[0.5171785, 0.23878628, 0.24403522],
                          [0.22141451, 0.64102704, 0.13755846],
                          [0.16751315, 0.18256843, 0.64991843],
                          [0.27357372, 0.55201592, 0.17441036],
                          [0.65745193, 0.26062899, 0.08191907]])]

    for i in range(len(y_actual)):
        assert_almost_equal(y_result[i], y_actual[i])


def test_multi_output_classification_sample_weights():
    # weighted classifier
    Xw = [[1, 2, 3], [4, 5, 6]]
    yw = [[3, 2], [2, 3]]
    w = np.asarray([2., 1.])
    forest = RandomForestClassifier(n_estimators=10, random_state=1)
    clf_w = MultiOutputClassifier(forest)
    clf_w.fit(Xw, yw, w)

    # unweighted, but with repeated samples
    X = [[1, 2, 3], [1, 2, 3], [4, 5, 6]]
    y = [[3, 2], [3, 2], [2, 3]]
    forest = RandomForestClassifier(n_estimators=10, random_state=1)
    clf = MultiOutputClassifier(forest)
    clf.fit(X, y)

    X_test = [[1.5, 2.5, 3.5], [3.5, 4.5, 5.5]]
    assert_almost_equal(clf.predict(X_test), clf_w.predict(X_test))


# 0.23. warning about tol not having its correct default value.
@pytest.mark.filterwarnings('ignore:max_iter and tol parameters have been')
def test_multi_output_classification_partial_fit_sample_weights():
    # weighted classifier
    Xw = [[1, 2, 3], [4, 5, 6], [1.5, 2.5, 3.5]]
    yw = [[3, 2], [2, 3], [3, 2]]
    w = np.asarray([2., 1., 1.])
    sgd_linear_clf = SGDClassifier(random_state=1, max_iter=20)
    clf_w = MultiOutputClassifier(sgd_linear_clf)
    clf_w.fit(Xw, yw, w)

    # unweighted, but with repeated samples
    X = [[1, 2, 3], [1, 2, 3], [4, 5, 6], [1.5, 2.5, 3.5]]
    y = [[3, 2], [3, 2], [2, 3], [3, 2]]
    sgd_linear_clf = SGDClassifier(random_state=1, max_iter=20)
    clf = MultiOutputClassifier(sgd_linear_clf)
    clf.fit(X, y)
    X_test = [[1.5, 2.5, 3.5]]
    assert_array_almost_equal(clf.predict(X_test), clf_w.predict(X_test))


def test_multi_output_exceptions():
    # NotFittedError when fit is not done but score, predict and
    # and predict_proba are called
    moc = MultiOutputClassifier(LinearSVC(random_state=0))
    assert_raises(NotFittedError, moc.predict, y)
    assert_raises(NotFittedError, moc.predict_proba, y)
    assert_raises(NotFittedError, moc.score, X, y)
    # ValueError when number of outputs is different
    # for fit and score
    y_new = np.column_stack((y1, y2))
    moc.fit(X, y)
    assert_raises(ValueError, moc.score, X, y_new)
    # ValueError when y is continuous
    assert_raise_message(ValueError, "Unknown label type", moc.fit, X, X[:, 1])


def generate_multilabel_dataset_with_correlations():
    # Generate a multilabel data set from a multiclass dataset as a way of
    # by representing the integer number of the original class using a binary
    # encoding.
    X, y = make_classification(n_samples=1000,
                               n_features=100,
                               n_classes=16,
                               n_informative=10,
                               random_state=0)

    Y_multi = np.array([[int(yyy) for yyy in format(yy, '#06b')[2:]]
                        for yy in y])
    return X, Y_multi


def test_classifier_chain_fit_and_predict_with_linear_svc():
    # Fit classifier chain and verify predict performance using LinearSVC
    X, Y = generate_multilabel_dataset_with_correlations()
    classifier_chain = ClassifierChain(LinearSVC())
    classifier_chain.fit(X, Y)

    Y_pred = classifier_chain.predict(X)
    assert Y_pred.shape == Y.shape

    Y_decision = classifier_chain.decision_function(X)

    Y_binary = (Y_decision >= 0)
    assert_array_equal(Y_binary, Y_pred)
    assert not hasattr(classifier_chain, 'predict_proba')


def test_classifier_chain_fit_and_predict_with_sparse_data():
    # Fit classifier chain with sparse data
    X, Y = generate_multilabel_dataset_with_correlations()
    X_sparse = sp.csr_matrix(X)

    classifier_chain = ClassifierChain(LogisticRegression())
    classifier_chain.fit(X_sparse, Y)
    Y_pred_sparse = classifier_chain.predict(X_sparse)

    classifier_chain = ClassifierChain(LogisticRegression())
    classifier_chain.fit(X, Y)
    Y_pred_dense = classifier_chain.predict(X)

    assert_array_equal(Y_pred_sparse, Y_pred_dense)


def test_classifier_chain_vs_independent_models():
    # Verify that an ensemble of classifier chains (each of length
    # N) can achieve a higher Jaccard similarity score than N independent
    # models
    X, Y = generate_multilabel_dataset_with_correlations()
    X_train = X[:600, :]
    X_test = X[600:, :]
    Y_train = Y[:600, :]
    Y_test = Y[600:, :]

    ovr = OneVsRestClassifier(LogisticRegression())
    ovr.fit(X_train, Y_train)
    Y_pred_ovr = ovr.predict(X_test)

    chain = ClassifierChain(LogisticRegression())
    chain.fit(X_train, Y_train)
    Y_pred_chain = chain.predict(X_test)

    assert (jaccard_score(Y_test, Y_pred_chain, average='samples') >
                   jaccard_score(Y_test, Y_pred_ovr, average='samples'))


def test_base_chain_fit_and_predict():
    # Fit base chain and verify predict performance
    X, Y = generate_multilabel_dataset_with_correlations()
    chains = [RegressorChain(Ridge()),
              ClassifierChain(LogisticRegression())]
    for chain in chains:
        chain.fit(X, Y)
        Y_pred = chain.predict(X)
        assert Y_pred.shape == Y.shape
        assert ([c.coef_.size for c in chain.estimators_] ==
                     list(range(X.shape[1], X.shape[1] + Y.shape[1])))

    Y_prob = chains[1].predict_proba(X)
    Y_binary = (Y_prob >= .5)
    assert_array_equal(Y_binary, Y_pred)

    assert isinstance(chains[1], ClassifierMixin)


def test_base_chain_fit_and_predict_with_sparse_data_and_cv():
    # Fit base chain with sparse data cross_val_predict
    X, Y = generate_multilabel_dataset_with_correlations()
    X_sparse = sp.csr_matrix(X)
    base_chains = [ClassifierChain(LogisticRegression(), cv=3),
                   RegressorChain(Ridge(), cv=3)]
    for chain in base_chains:
        chain.fit(X_sparse, Y)
        Y_pred = chain.predict(X_sparse)
        assert Y_pred.shape == Y.shape


def test_base_chain_random_order():
    # Fit base chain with random order
    X, Y = generate_multilabel_dataset_with_correlations()
    for chain in [ClassifierChain(LogisticRegression()),
                  RegressorChain(Ridge())]:
        chain_random = clone(chain).set_params(order='random', random_state=42)
        chain_random.fit(X, Y)
        chain_fixed = clone(chain).set_params(order=chain_random.order_)
        chain_fixed.fit(X, Y)
        assert_array_equal(chain_fixed.order_, chain_random.order_)
        assert list(chain_random.order) != list(range(4))
        assert len(chain_random.order_) == 4
        assert len(set(chain_random.order_)) == 4
        # Randomly ordered chain should behave identically to a fixed order
        # chain with the same order.
        for est1, est2 in zip(chain_random.estimators_,
                              chain_fixed.estimators_):
            assert_array_almost_equal(est1.coef_, est2.coef_)


def test_base_chain_crossval_fit_and_predict():
    # Fit chain with cross_val_predict and verify predict
    # performance
    X, Y = generate_multilabel_dataset_with_correlations()

    for chain in [ClassifierChain(LogisticRegression()),
                  RegressorChain(Ridge())]:
        chain.fit(X, Y)
        chain_cv = clone(chain).set_params(cv=3)
        chain_cv.fit(X, Y)
        Y_pred_cv = chain_cv.predict(X)
        Y_pred = chain.predict(X)

        assert Y_pred_cv.shape == Y_pred.shape
        assert not np.all(Y_pred == Y_pred_cv)
        if isinstance(chain, ClassifierChain):
            assert jaccard_score(Y, Y_pred_cv, average='samples') > .4
        else:
            assert mean_squared_error(Y, Y_pred_cv) < .25