def test_quantile_bins():
bins = 4
random_state = np.random.RandomState(0)
x = random_state.normal(0, 1, size=1000)
myBucketer = QuantileBucketer(bin_count=bins)
with pytest.raises(NotFittedError):
myBucketer.compute([1, 2])
myBucketer.fit(x)
assert len(myBucketer.counts_) == bins
assert np.array_equal(myBucketer.counts_, np.array([250, 250, 250, 250]))
assert len(myBucketer.boundaries_) == bins + 1
np.testing.assert_array_almost_equal(myBucketer.boundaries_,
np.array([-3.0, -0.7, -0.1, 0.6,
2.8]),
decimal=1)
# test static method
counts, boundaries = QuantileBucketer(bin_count=bins).quantile_bins(
x, bins)
assert np.array_equal(myBucketer.counts_, counts)
np.testing.assert_array_almost_equal(myBucketer.boundaries_, boundaries)
# test inf edges
counts, boundaries = QuantileBucketer(bin_count=bins).quantile_bins(
x, bins, inf_edges=True)
assert boundaries[0] == -np.inf
assert boundaries[-1] == np.inf
assert repr(myBucketer).startswith('QuantileBucketer')
def test_quantile_with_unique_values():
np.random.seed(42)
dist_0_1 = np.random.uniform(size=20)
dist_peak_at_0 = np.zeros(shape=20)
skewed_dist = np.hstack((dist_0_1, dist_peak_at_0))
actual_out = QuantileBucketer(10).quantile_bins(skewed_dist, 10)
expected_out = (np.array([20, 4, 4, 4, 4, 4]),
np.array([
0., 0.01894458, 0.23632033, 0.42214475, 0.60977678,
0.67440958, 0.99940487
]))
assert (actual_out[0] == expected_out[0]).all()
def _target_rate_plot(self,
feature,
bins=10,
type_binning="simple",
ax=None):
"""
Plots the distributions of the specific features, as well as the target rate as function of the feature.
Args:
feature (str or int):
Feature for which to create target rate plot.
bins (int or list[float]), optional:
Number of bins or boundaries of desired bins in list.
type_binning ({'simple', 'agglomerative', 'quantile'}, optional):
Type of binning strategy used to create bins.
ax (matplotlib.pyplot.axes, optional):
Optional axis on which to draw plot.
Returns:
(list[float], matplotlib.pyplot.axes, float):
Tuple of boundaries of bins used, axis on which plot is drawn, total ratio of target (positive over
negative).
"""
x, y, shap_val = self._get_X_y_shap_with_q_cut(feature=feature)
# Create bins if not explicitly supplied
if type(bins) is int:
if type_binning == "simple":
counts, bins = SimpleBucketer.simple_bins(x, bins)
elif type_binning == "agglomerative":
counts, bins = AgglomerativeBucketer.agglomerative_clustering_binning(
x, bins)
elif type_binning == "quantile":
counts, bins = QuantileBucketer.quantile_bins(x, bins)
# Determine bin for datapoints
bins[-1] = bins[-1] + 1
indices = np.digitize(x, bins)
# Create dataframe with binned data
dfs = pd.DataFrame({
feature: x,
"y": y,
"bin_index": pd.Series(indices, index=x.index)
}).groupby("bin_index", as_index=True)
# Extract target ratio and mean feature value
target_ratio = dfs["y"].mean()
x_vals = dfs[feature].mean()
# Plot target rate
ax.hist(x, bins=bins, lw=2, alpha=0.4)
ax.set_ylabel("Counts")
ax2 = ax.twinx()
ax2.plot(x_vals, target_ratio, color="red")
ax2.set_ylabel("Target rate", color="red", fontsize=12)
ax2.set_xlim(x.min(), x.max())
ax.set_xlabel(f'{feature} feature values')
return bins, ax, target_ratio
def test_compute():
x = np.arange(10)
bins = 5
myBucketer = QuantileBucketer(bins)
x_new = x
with pytest.raises(NotFittedError):
assert myBucketer.compute(x_new)
myBucketer.fit(x)
assert len(myBucketer.compute(x_new)) == bins
np.testing.assert_array_equal(myBucketer.counts_,
myBucketer.compute(x_new))
np.testing.assert_array_equal(myBucketer.counts_,
myBucketer.fit_compute(x_new))
x_new = x + 100
np.testing.assert_array_equal(np.array([0, 0, 0, 0, 0]),
myBucketer.compute(x_new))
x_new = x - 100
np.testing.assert_array_equal(np.array([10, 0, 0, 0, 0]),
myBucketer.compute(x_new))
x_new = [1, 1, 1, 4, 4, 7]
np.testing.assert_array_equal(np.array([3, 0, 2, 1, 0]),
myBucketer.compute(x_new))