def test_invalid_encode_option():
est = KBinsDiscretizer(n_bins=[2, 3, 3, 3], encode='invalid-encode')
err_msg = (r"Valid options for 'encode' are "
r"\('onehot', 'onehot-dense', 'ordinal'\). "
r"Got encode='invalid-encode' instead.")
with pytest.raises(ValueError, match=err_msg):
est.fit(X)
def test_invalid_strategy_option():
est = KBinsDiscretizer(n_bins=[2, 3, 3, 3], strategy='invalid-strategy')
err_msg = (r"Valid options for 'strategy' are "
r"\('uniform', 'quantile', 'kmeans'\). "
r"Got strategy='invalid-strategy' instead.")
with pytest.raises(ValueError, match=err_msg):
est.fit(X)
def test_permutation_importance_equivalence_array_dataframe(n_jobs):
# This test checks that the column shuffling logic has the same behavior
# both a dataframe and a simple numpy array.
pd = pytest.importorskip('pandas')
# regression test to make sure that sequential and parallel calls will
# output the same results.
X, y = make_regression(n_samples=100, n_features=5, random_state=0)
X_df = pd.DataFrame(X)
# Add a categorical feature that is statistically linked to y:
binner = KBinsDiscretizer(n_bins=3, encode="ordinal")
cat_column = binner.fit_transform(y.reshape(-1, 1))
# Concatenate the extra column to the numpy array: integers will be
# cast to float values
X = np.hstack([X, cat_column])
assert X.dtype.kind == "f"
# Insert extra column as a non-numpy-native dtype (while keeping backward
# compat for old pandas versions):
if hasattr(pd, "Categorical"):
cat_column = pd.Categorical(cat_column.ravel())
else:
cat_column = cat_column.ravel()
new_col_idx = len(X_df.columns)
X_df[new_col_idx] = cat_column
assert X_df[new_col_idx].dtype == cat_column.dtype
# Stich an aribtrary index to the dataframe:
X_df.index = np.arange(len(X_df)).astype(str)
rf = RandomForestRegressor(n_estimators=5, max_depth=3, random_state=0)
rf.fit(X, y)
n_repeats = 3
importance_array = permutation_importance(rf,
X,
y,
n_repeats=n_repeats,
random_state=0,
n_jobs=n_jobs)
# First check that the problem is structured enough and that the model is
# complex enough to not yield trivial, constant importances:
imp_min = importance_array['importances'].min()
imp_max = importance_array['importances'].max()
assert imp_max - imp_min > 0.3
# Now check that importances computed on dataframe matche the values
# of those computed on the array with the same data.
importance_dataframe = permutation_importance(rf,
X_df,
y,
n_repeats=n_repeats,
random_state=0,
n_jobs=n_jobs)
assert_allclose(importance_array['importances'],
importance_dataframe['importances'])
def test_transform_outside_fit_range(strategy):
X = np.array([0, 1, 2, 3])[:, None]
kbd = KBinsDiscretizer(n_bins=4, strategy=strategy, encode='ordinal')
kbd.fit(X)
X2 = np.array([-2, 5])[:, None]
X2t = kbd.transform(X2)
assert_array_equal(X2t.max(axis=0) + 1, kbd.n_bins_)
assert_array_equal(X2t.min(axis=0), [0])
def test_transform_1d_behavior():
X = np.arange(4)
est = KBinsDiscretizer(n_bins=2)
with pytest.raises(ValueError):
est.fit(X)
est = KBinsDiscretizer(n_bins=2)
est.fit(X.reshape(-1, 1))
with pytest.raises(ValueError):
est.transform(X)
def test_fit_transform_n_bins_array(strategy, expected):
est = KBinsDiscretizer(n_bins=[2, 3, 3, 3], encode='ordinal',
strategy=strategy).fit(X)
assert_array_equal(expected, est.transform(X))
# test the shape of bin_edges_
n_features = np.array(X).shape[1]
assert est.bin_edges_.shape == (n_features, )
for bin_edges, n_bins in zip(est.bin_edges_, est.n_bins_):
assert bin_edges.shape == (n_bins + 1, )
def test_percentile_numeric_stability():
X = np.array([0.05, 0.05, 0.95]).reshape(-1, 1)
bin_edges = np.array([0.05, 0.23, 0.41, 0.59, 0.77, 0.95])
Xt = np.array([0, 0, 4]).reshape(-1, 1)
kbd = KBinsDiscretizer(n_bins=10, encode='ordinal',
strategy='quantile')
msg = ("Bins whose width are too small (i.e., <= 1e-8) in feature 0 "
"are removed. Consider decreasing the number of bins.")
assert_warns_message(UserWarning, msg, kbd.fit, X)
assert_array_almost_equal(kbd.bin_edges_[0], bin_edges)
assert_array_almost_equal(kbd.transform(X), Xt)
def test_overwrite():
X = np.array([0, 1, 2, 3])[:, None]
X_before = X.copy()
est = KBinsDiscretizer(n_bins=3, encode="ordinal")
Xt = est.fit_transform(X)
assert_array_equal(X, X_before)
Xt_before = Xt.copy()
Xinv = est.inverse_transform(Xt)
assert_array_equal(Xt, Xt_before)
assert_array_equal(Xinv, np.array([[0.5], [1.5], [2.5], [2.5]]))
def test_same_min_max(strategy):
warnings.simplefilter("always")
X = np.array([[1, -2],
[1, -1],
[1, 0],
[1, 1]])
est = KBinsDiscretizer(strategy=strategy, n_bins=3, encode='ordinal')
assert_warns_message(UserWarning,
"Feature 0 is constant and will be replaced "
"with 0.", est.fit, X)
assert est.n_bins_[0] == 1
# replace the feature with zeros
Xt = est.transform(X)
assert_array_equal(Xt[:, 0], np.zeros(X.shape[0]))
def test_redundant_bins(strategy, expected_bin_edges):
X = [[0], [0], [0], [0], [3], [3]]
kbd = KBinsDiscretizer(n_bins=3, strategy=strategy)
msg = ("Bins whose width are too small (i.e., <= 1e-8) in feature 0 "
"are removed. Consider decreasing the number of bins.")
assert_warns_message(UserWarning, msg, kbd.fit, X)
assert_array_almost_equal(kbd.bin_edges_[0], expected_bin_edges)
def test_numeric_stability(i):
X_init = np.array([2., 4., 6., 8., 10.]).reshape(-1, 1)
Xt_expected = np.array([0, 0, 1, 1, 1]).reshape(-1, 1)
# Test up to discretizing nano units
X = X_init / 10**i
Xt = KBinsDiscretizer(n_bins=2, encode='ordinal').fit_transform(X)
assert_array_equal(Xt_expected, Xt)
def test_invalid_n_bins():
est = KBinsDiscretizer(n_bins=1)
err_msg = ("KBinsDiscretizer received an invalid "
"number of bins. Received 1, expected at least 2.")
with pytest.raises(ValueError, match=err_msg):
est.fit_transform(X)
est = KBinsDiscretizer(n_bins=1.1)
err_msg = ("KBinsDiscretizer received an invalid "
"n_bins type. Received float, expected int.")
with pytest.raises(ValueError, match=err_msg):
est.fit_transform(X)
def test_nonuniform_strategies(
strategy, expected_2bins, expected_3bins, expected_5bins):
X = np.array([0, 0.5, 2, 3, 9, 10]).reshape(-1, 1)
# with 2 bins
est = KBinsDiscretizer(n_bins=2, strategy=strategy, encode='ordinal')
Xt = est.fit_transform(X)
assert_array_equal(expected_2bins, Xt.ravel())
# with 3 bins
est = KBinsDiscretizer(n_bins=3, strategy=strategy, encode='ordinal')
Xt = est.fit_transform(X)
assert_array_equal(expected_3bins, Xt.ravel())
# with 5 bins
est = KBinsDiscretizer(n_bins=5, strategy=strategy, encode='ordinal')
Xt = est.fit_transform(X)
assert_array_equal(expected_5bins, Xt.ravel())
def test_encode_options():
est = KBinsDiscretizer(n_bins=[2, 3, 3, 3],
encode='ordinal').fit(X)
Xt_1 = est.transform(X)
est = KBinsDiscretizer(n_bins=[2, 3, 3, 3],
encode='onehot-dense').fit(X)
Xt_2 = est.transform(X)
assert not sp.issparse(Xt_2)
assert_array_equal(OneHotEncoder(
categories=[np.arange(i) for i in [2, 3, 3, 3]],
sparse=False)
.fit_transform(Xt_1), Xt_2)
est = KBinsDiscretizer(n_bins=[2, 3, 3, 3],
encode='onehot').fit(X)
Xt_3 = est.transform(X)
assert sp.issparse(Xt_3)
assert_array_equal(OneHotEncoder(
categories=[np.arange(i) for i in [2, 3, 3, 3]],
sparse=True)
.fit_transform(Xt_1).toarray(),
Xt_3.toarray())
def make_missing_value_data(n_samples=int(1e4), seed=0):
rng = np.random.RandomState(seed)
X, y = make_regression(n_samples=n_samples,
n_features=4,
random_state=rng)
# Pre-bin the data to ensure a deterministic handling by the 2
# strategies and also make it easier to insert np.nan in a structured
# way:
X = KBinsDiscretizer(n_bins=42, encode="ordinal").fit_transform(X)
# First feature has missing values completely at random:
rnd_mask = rng.rand(X.shape[0]) > 0.9
X[rnd_mask, 0] = np.nan
# Second and third features have missing values for extreme values
# (censoring missingness):
low_mask = X[:, 1] == 0
X[low_mask, 1] = np.nan
high_mask = X[:, 2] == X[:, 2].max()
X[high_mask, 2] = np.nan
# Make the last feature nan pattern very informative:
y_max = np.percentile(y, 70)
y_max_mask = y >= y_max
y[y_max_mask] = y_max
X[y_max_mask, 3] = np.nan
# Check that there is at least one missing value in each feature:
for feature_idx in range(X.shape[1]):
assert any(np.isnan(X[:, feature_idx]))
# Let's use a test set to check that the learned decision function is
# the same as evaluated on unseen data. Otherwise it could just be the
# case that we find two independent ways to overfit the training set.
return train_test_split(X, y, random_state=rng)
def test_invalid_n_bins_array():
# Bad shape
n_bins = np.full((2, 4), 2.)
est = KBinsDiscretizer(n_bins=n_bins)
err_msg = r"n_bins must be a scalar or array of shape \(n_features,\)."
with pytest.raises(ValueError, match=err_msg):
est.fit_transform(X)
# Incorrect number of features
n_bins = [1, 2, 2]
est = KBinsDiscretizer(n_bins=n_bins)
err_msg = r"n_bins must be a scalar or array of shape \(n_features,\)."
with pytest.raises(ValueError, match=err_msg):
est.fit_transform(X)
# Bad bin values
n_bins = [1, 2, 2, 1]
est = KBinsDiscretizer(n_bins=n_bins)
err_msg = ("KBinsDiscretizer received an invalid number of bins "
"at indices 0, 3. Number of bins must be at least 2, "
"and must be an int.")
with pytest.raises(ValueError, match=err_msg):
est.fit_transform(X)
# Float bin values
n_bins = [2.1, 2, 2.1, 2]
est = KBinsDiscretizer(n_bins=n_bins)
err_msg = ("KBinsDiscretizer received an invalid number of bins "
"at indices 0, 2. Number of bins must be at least 2, "
"and must be an int.")
with pytest.raises(ValueError, match=err_msg):
est.fit_transform(X)
def test_valid_n_bins():
KBinsDiscretizer(n_bins=2).fit_transform(X)
KBinsDiscretizer(n_bins=np.array([2])[0]).fit_transform(X)
assert KBinsDiscretizer(n_bins=2).fit(X).n_bins_.dtype == np.dtype(np.int)
def test_fit_transform(strategy, expected):
est = KBinsDiscretizer(n_bins=3, encode='ordinal', strategy=strategy)
est.fit(X)
assert_array_equal(expected, est.transform(X))
def test_invalid_n_features():
est = KBinsDiscretizer(n_bins=3).fit(X)
bad_X = np.arange(25).reshape(5, -1)
err_msg = "Incorrect number of features. Expecting 4, received 5"
with pytest.raises(ValueError, match=err_msg):
est.transform(bad_X)
def test_inverse_transform(strategy, encode, expected_inv):
kbd = KBinsDiscretizer(n_bins=3, strategy=strategy, encode=encode)
Xt = kbd.fit_transform(X)
Xinv = kbd.inverse_transform(Xt)
assert_array_almost_equal(expected_inv, Xinv)
