def test_class_prediction_error_quickmethod_X_test_only(self):
"""
Test the ClassPredictionError quickmethod
"""
X, y = load_occupancy(return_dataset=True).to_numpy()
X_train, X_test, y_train, y_test = tts(X,
y,
test_size=0.2,
shuffle=True,
random_state=42)
fig = plt.figure()
ax = fig.add_subplot()
clf = LinearSVC(random_state=42)
with pytest.raises(
YellowbrickValueError,
match="must specify both X_test and y_test or neither",
):
class_prediction_error(clf,
X_train=X_train,
y_train=y_train,
X_test=X_test,
ax=ax,
show=False)
def test_integrated_radviz_pandas_classes_features(self):
"""
Test RadViz with classes and features specified using Pandas
"""
# Load the data from the fixture
data = load_occupancy(return_dataset=True)
X, y = data.to_pandas()
features = ["temperature", "relative humidity", "light"]
classes = [
k
for k, _ in sorted(data.meta["labels"].items(), key=lambda i: i[1])
]
assert isinstance(X, pd.DataFrame)
assert isinstance(y, pd.Series)
# Filter the dataset to make sure it's not just class names
X = X[features]
y = y.astype(int)
# Test the visualizer
visualizer = RadViz(features=features, classes=classes)
visualizer.fit_transform(X, y)
visualizer.finalize()
self.assert_images_similar(visualizer, tol=0.1)
def test_integrated_radviz_numpy_classes_features(self):
"""
Test RadViz with classes and features specified using numpy
"""
# Load the data from the fixture
data = load_occupancy(return_dataset=True)
X, y = data.to_numpy()
features = data.meta["features"][0:3]
classes = [
k
for k, _ in sorted(data.meta["labels"].items(), key=lambda i: i[1])
]
assert isinstance(X, np.ndarray)
assert isinstance(y, np.ndarray)
# Filter the dataset to make sure it's not just class names
X = X[:, :3]
y = y.astype(int)
# Test the visualizer
visualizer = RadViz(features=features, classes=classes)
visualizer.fit_transform(X, y)
visualizer.finalize()
self.assert_images_similar(visualizer, tol=0.1)
def balance():
X, y = load_occupancy()
_, _, y_train, y_test = tts(X, y, test_size=0.2)
oz = ClassBalance(ax=newfig(), labels=["unoccupied", "occupied"])
oz.fit(y_train, y_test)
savefig(oz, "class_balance")
def classreport():
X, y = load_occupancy()
X_train, X_test, y_train, y_test = tts(X, y, test_size=0.2)
oz = ClassificationReport(GaussianNB(), support=True, ax=newfig())
oz.fit(X_train, y_train)
oz.score(X_test, y_test)
savefig(oz, "classification_report")
def test_pandas_integration(self):
"""
Test the precision_recall_curve with Pandas dataframes
"""
X, y = load_occupancy(return_dataset=True).to_pandas()
model = DecisionTreeClassifier(random_state=14)
X_train, X_test, y_train, y_test = tts(
X, y, test_size=0.2, shuffle=True, random_state=555
)
oz = PrecisionRecallCurve(
model,
per_class=True,
micro=False,
fill_area=False,
iso_f1_curves=True,
ap_score=False,
classes=["unoccupied", "occupied"],
)
oz.fit(X_train, y_train)
oz.score(X_test, y_test)
oz.finalize()
self.assert_images_similar(oz, tol=5.0)
def test_class_prediction_error_quickmethod_X_test_and_y_test(self):
"""
Test the ClassPredictionError quickmethod
"""
X, y = load_occupancy(return_dataset=True).to_numpy()
X_train, X_test, y_train, y_test = tts(X,
y,
test_size=0.2,
shuffle=True,
random_state=42)
fig = plt.figure()
ax = fig.add_subplot()
clf = LinearSVC(random_state=42)
viz = class_prediction_error(clf,
X_train=X_train,
y_train=y_train,
X_test=X_test,
y_test=y_test,
ax=ax,
show=False)
# Not sure why the tolerance must be so high for this
# Failing on travis with RMS 9.544
# AppVeyor and Linux conda fail due to non-text-based differences: RMS 12.961
self.assert_images_similar(viz, tol=13, windows_tol=13)
def test_missing_test_data_in_quick_method(self):
"""
Test quick method when test data is missing.
"""
X, y = load_occupancy(return_dataset=True).to_numpy()
X_train, X_test, y_train, y_test = tts(X,
y,
test_size=0.2,
shuffle=True,
random_state=55555)
emsg = "both X_test and y_test are required if one is specified"
with pytest.raises(YellowbrickValueError, match=emsg):
precision_recall_curve(RandomForestClassifier(),
X_train,
y_train,
y_test=y_test,
show=False)
with pytest.raises(YellowbrickValueError, match=emsg):
precision_recall_curve(RandomForestClassifier(),
X_train,
y_train,
X_test,
show=False)
def test_rocauc_quickmethod(self):
"""
Test the ROCAUC quick method
"""
X, y = load_occupancy(return_dataset=True).to_numpy()
model = DecisionTreeClassifier()
# TODO: image comparison of the quick method
roc_auc(model, X, y)
def test_rank2d_quick_method(self):
"""
Test Rank2D quick method
"""
X, y = load_occupancy()
oz = rank2d(X, y, algorithm="spearman", colormap="RdYlGn_r")
assert isinstance(oz, Rank2D)
self.assert_images_similar(oz, tol=0.1)
def test_rocauc_quickmethod(self):
"""
Test the ROCAUC quick method
"""
X, y = load_occupancy(return_dataset=True).to_numpy()
model = LogisticRegression()
# compare the images
visualizer = roc_auc(model, X, y, show=False)
self.assert_images_similar(visualizer)
def test_parallel_coordinates_quickmethod(self):
"""
Test the quick method producing a valid visualization
"""
X, y = load_occupancy(return_dataset=True).to_numpy()
# Compare the images
# Use only the first 100 samples so the test will run faster
visualizer = parallel_coordinates(X, y, sample=100, show=False)
self.assert_images_similar(visualizer)
def test_quick_method(self):
"""
Test the quick method with producing a valid visualization
"""
data = load_occupancy(return_dataset=True)
_, y = data.to_numpy()
visualizer = balanced_binning_reference(y, show=False)
assert isinstance(visualizer, BalancedBinningReference)
self.assert_images_similar(visualizer, tol=0.5)
def manifold(dataset, manifold):
if dataset == "concrete":
X, y = load_concrete()
elif dataset == "occupancy":
X, y = load_occupancy()
else:
raise ValueError("unknown dataset")
oz = Manifold(manifold=manifold, ax=newfig())
oz.fit_transform(X, y)
savefig(oz, "{}_{}_manifold".format(dataset, manifold))
def test_integrated_scatter(self):
"""
Test scatter on the real, occupancy data set
"""
# Load the data from the fixture
X, y = load_occupancy(return_dataset=True).to_numpy()
# Test the visualizer
features = ["temperature", "relative humidity"]
visualizer = ScatterViz(features=features)
visualizer.fit_transform_show(X[:, :2], y)
def test_classes_greater_than_indices(self):
"""
A model error should be raised when there are more classes in fit than score
"""
X, y = load_occupancy(return_dataset=True).to_numpy()
classes = ["unoccupied", "occupied", "partytime"]
model = LinearSVC(random_state=42)
model.fit(X, y)
with pytest.raises(ModelError):
visualizer = ClassPredictionError(model, classes=classes)
visualizer.score(X, y)
def test_classes_less_than_indices(self):
"""
Assert error when there is an attempt to filter classes
"""
X, y = load_occupancy(return_dataset=True).to_numpy()
classes = ["unoccupied"]
model = LinearSVC(random_state=42)
model.fit(X, y)
with pytest.raises(NotImplementedError):
visualizer = ClassPredictionError(model, classes=classes)
visualizer.score(X, y)
def test_pandas_bins(self):
"""
Test Histogram on a Pandas Dataframe
"""
# Load the data from the fixture
data = load_occupancy(return_dataset=True)
X, y = data.to_pandas()
visualizer = BalancedBinningReference()
visualizer.fit(y)
visualizer.finalize()
self.assert_images_similar(visualizer, tol=0.5)
def test_integrated_scatter_with_pandas(self):
"""
Test scatterviz on the real, occupancy data set with pandas
"""
# Load the data from the fixture
# Load the data from the fixture
X, y = load_occupancy(return_dataset=True).to_pandas()
# Test the visualizer
features = ["temperature", "relative humidity"]
visualizer = ScatterViz(features=features)
visualizer.fit_transform_poof(X, y)
def test_score_returns_score(self):
"""
Test that ClassPredictionError score() returns a score between 0 and 1
"""
X, y = load_occupancy(return_dataset=True).to_numpy()
# Create and fit the visualizer
visualizer = ClassPredictionError(LinearSVC(random_state=42))
visualizer.fit(X, y)
# Score the visualizer
s = visualizer.score(X, y)
assert 0 <= s <= 1
def test_scatter_quick_method(self):
"""
Test scatter quick method on the real, occupancy data set
"""
# Load the data from the fixture
X, y = load_occupancy(return_dataset=True).to_numpy()
# Test the visualizer
features = ["temperature", "relative humidity"]
viz = scatterviz(X[:, :2], y=y, ax=None, features=features)
# test that is returns a matplotlib obj with axes
assert isinstance(viz, ScatterVisualizer)
def test_numpy_occupancy_balance(self):
"""
Test NumPy arrays with string target in balance mode
"""
data = load_occupancy(return_dataset=True)
X, y = data.to_numpy()
# Create and fit the visualizer
oz = ClassBalance()
assert oz.fit(y) is oz
# oz.finalize()
self.assert_images_similar(oz)
def select_features_example(
algorithm="isomap",
path="images/occupancy_select_k_best_isomap_manifold.png",
**kwargs):
_, ax = plt.subplots(figsize=(9, 6))
model = Pipeline([
("selectk", SelectKBest(k=3, score_func=f_classif)),
("viz", Manifold(ax=ax, manifold=algorithm, **kwargs)),
])
X, y = load_occupancy()
model.fit(X, y)
model.named_steps["viz"].poof(outpath=path)
def test_integrated_radviz_with_pandas(self):
"""
Test RadViz with Pandas on the occupancy dataset
"""
data = load_occupancy(return_dataset=True)
X, y = data.to_pandas()
assert isinstance(X, pd.DataFrame)
assert isinstance(y, pd.Series)
# Test the visualizer
visualizer = RadViz()
visualizer.fit_transform_poof(X, y)
self.assert_images_similar(visualizer, tol=0.1)
def test_integrated_radviz_with_numpy(self):
"""
Test RadViz with numpy on the occupancy dataset
"""
data = load_occupancy(return_dataset=True)
X, y = data.to_numpy()
assert isinstance(X, np.ndarray)
assert isinstance(y, np.ndarray)
# Test the visualizer
visualizer = RadViz()
visualizer.fit_transform_poof(X, y)
self.assert_images_similar(visualizer, tol=0.1)
def test_quick_method_with_test_set(self):
"""
Test quick method when both train and test data is supplied
"""
X, y = load_occupancy(return_dataset=True).to_numpy()
X_train, X_test, y_train, y_test = tts(
X, y, test_size=0.2, shuffle=True, random_state=555
)
viz = precision_recall_curve(
RandomForestClassifier(random_state=72), X_train, y_train, X_test, y_test
)
self.assert_images_similar(viz)
def test_pandas_occupancy_compare(self):
"""
Test pandas data frame with string target in compare mode
"""
data = load_occupancy(return_dataset=True)
X, y = data.to_pandas()
_, _, y_train, y_test = tts(X, y, test_size=0.4, random_state=2242)
# Create and fit the visualizer
oz = ClassBalance()
assert oz.fit(y_train, y_test) is oz
# oz.finalize()
self.assert_images_similar(oz, tol=0.5) # w/o tol fails with RMS 0.433
def test_numpy_occupancy_compare(self):
"""
Test NumPy arrays with string target in compare mode
"""
data = load_occupancy(return_dataset=True)
X, y = data.to_numpy()
_, _, y_train, y_test = tts(X, y, test_size=0.4, random_state=2242)
# Create and fit the visualizer
oz = ClassBalance()
assert oz.fit(y_train, y_test) is oz
# oz.finalize()
self.assert_images_similar(oz)
def test_stack_param_incorrectly_used_throws_error(self):
"""
Test incorrectly using stack param on a dataset with two classes which
does not return a coef_ array in the shape of (n_classes, n_features)
"""
X, y = load_occupancy()
viz = FeatureImportances(
LogisticRegression(solver="liblinear", random_state=222), stack=True
)
expected_error = "The model used does not return coef_ array"
with pytest.raises(YellowbrickValueError, match=expected_error):
viz.fit(X, y)
def test_pandas_integration(self):
"""
Assert no errors during class prediction error integration with Pandas
"""
X, y = load_occupancy(return_dataset=True).to_pandas()
classes = ["unoccupied", "occupied"]
model = LinearSVC(random_state=42)
model.fit(X, y)
visualizer = ClassPredictionError(model, classes=classes)
visualizer.score(X, y)
visualizer.finalize()
# AppVeyor and Linux conda fail due to non-text-based differences
# AppVeyor fails with RMS 13.161 - 13.289 (python - miniconda)
self.assert_images_similar(visualizer, tol=12.5, windows_tol=13.3)
def compare_class_balance(path="images/class_balance_compare.png"):
data = load_occupancy()
features = ["temperature", "relative_humidity", "light", "C02", "humidity"]
classes = ['unoccupied', 'occupied']
# Extract the numpy arrays from the data frame
X = data[features]
y = data["occupancy"]
# Create the train and test data
_, _, y_train, y_test = train_test_split(X, y, test_size=0.2)
# Instantiate the classification model and visualizer
visualizer = ClassBalance(labels=classes)
visualizer.fit(y_train, y_test)
return visualizer.poof(outpath=path)