def plot_hyperplane(clf, ax: Axes = None, interval: float = .05, alpha=.3,
colors=('r', 'b')) -> Line3DCollection:
"""
Plots the hyperplane of the model in an axes
:param clf: the classifier to use to find the hyperplane
:param ax: the axes to plot the hyperplane into
:param interval: the precision of the the hyperplane rendering.
:param alpha:
:param colors:
:return: the mesh of the created hyperplane that was added to the axes
"""
is_3d, ax = needed_axes(clf, ax)
interval = int(1 / interval)
# get the separating hyperplane
x_min, x_max = ax.get_xlim()
y_min, y_max = ax.get_ylim()
# create grid to evaluate model
xx = np.linspace(x_min, x_max, interval)
yy = np.linspace(y_min, y_max, interval)
if is_3d:
z_min, z_max = ax.get_zlim()
zz = np.linspace(z_min, z_max, interval)
yy, xx, zz = np.meshgrid(yy, xx, zz)
if hasattr(clf, "decision_function"):
z = clf.decision_function(np.c_[xx.ravel(), yy.ravel(), zz.ravel()])
elif hasattr(clf, "predict_proba"):
z = clf.predict_proba(np.c_[xx.ravel(), yy.ravel(), zz.ravel()])[:, 1]
else:
raise ValueError(
"The model passed in does not contain either the decision_function or the predict_proba functions.")
z = z.reshape(xx.shape)
vertices, faces, _, _ = measure.marching_cubes_lewiner(z, 0)
# Scale and transform to actual size of the interesting volume
vertices = vertices * [x_max - x_min, y_max - y_min, z_max - z_min] / interval
vertices += [x_min, y_min, z_min]
# and create a mesh to display
mesh = Line3DCollection(vertices[faces],
facecolor=colors, alpha=alpha)
ax.add_collection3d(mesh)
return mesh
else:
xx, yy = np.meshgrid(xx,
yy)
Z = clf.predict(np.c_[xx.ravel(), yy.ravel()])
Z = Z.reshape(xx.shape)
return ax.contourf(xx, yy, Z, 10, colors=colors, alpha=alpha)
def plot_hyperplane(clf, ax: Axes = None, interval: float = .05, alpha=.3,
colors: list = ('r', 'b')) -> Line3DCollection:
"""
Plots the hyperplane of the model in an axes
:param clf: the classifier to use to find the hyperplane
:param ax: the axes to plot the hyperplane into
:param interval: the precision of the the hyperplane rendering.
:return: the mesh of the created hyperplane that was added to the axes
"""
is_3d = False
if ax is None:
try:
clf.predict([[0, 0, 0]])
is_3d = True
ax = plt.gca(projection="3d")
except ValueError:
is_3d = False
ax = plt.gca()
elif isinstance(ax, Axes3D):
is_3d = True
interval = int(1 / interval)
# get the separating hyperplane
x_min, x_max = ax.get_xlim()
y_min, y_max = ax.get_ylim()
# create grid to evaluate model
xx = np.linspace(x_min, x_max, interval)
yy = np.linspace(y_min, y_max, interval)
if is_3d:
z_min, z_max = ax.get_zlim()
zz = np.linspace(z_min, z_max, interval)
yy, xx, zz = np.meshgrid(yy, xx, zz)
if hasattr(clf, "decision_function"):
z = clf.decision_function(np.c_[xx.ravel(), yy.ravel(), zz.ravel()])
elif hasattr(clf, "predict_proba"):
z = clf.predict_proba(np.c_[xx.ravel(), yy.ravel(), zz.ravel()])[:, 1]
z = z.reshape(xx.shape)
vertices, faces, _, _ = measure.marching_cubes(z, 0)
# Scale and transform to actual size of the interesting volume
vertices = vertices * [x_max - x_min, y_max - y_min, z_max - z_min] / interval
vertices += [x_min, y_min, z_min]
# and create a mesh to display
mesh = Line3DCollection(vertices[faces],
facecolor=colors, alpha=alpha)
ax.add_collection3d(mesh)
return mesh
else:
xx, yy = np.meshgrid(xx,
yy)
Z = clf.predict(np.c_[xx.ravel(), yy.ravel()])
Z = Z.reshape(xx.shape)
return ax.contourf(xx, yy, Z, 10, colors=colors, alpha=alpha)
