class TestKmeans(unittest.TestCase):
def setUp(self):
self.data = Table('iris')
new_domain = Domain(self.data.domain.attributes[:2])
self.data = Table(new_domain, self.data)
# self.centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans = Kmeans(self.data)
def test_k(self):
centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans.add_centroids(centroids)
self.assertEqual(self.kmeans.k, 3)
self.assertEqual(self.kmeans.k, len(self.kmeans.centroids))
def test_converged(self):
centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans.add_centroids(centroids)
self.assertFalse(self.kmeans.converged)
self.kmeans.step()
self.assertFalse(self.kmeans.converged)
# step not complete so false every odd state
self.kmeans.step()
# it is even step so maybe converged but it depends on example
# unable to test
self.kmeans.step()
self.assertFalse(self.kmeans.converged)
# check if false every not completed step
for i in range(self.kmeans.max_iter // 2 + 1):
self.kmeans.step()
self.kmeans.step()
self.assertFalse(self.kmeans.converged)
# it converged because of max iter
self.kmeans.step()
self.assertTrue(self.kmeans.converged)
def test_centroids_belonging_points(self):
centroids = [[5.2, 3.6]]
self.kmeans.add_centroids(centroids)
# if only one cluster all points in 0th element of first dimension
np.testing.assert_equal(self.kmeans.centroids_belonging_points,
np.array([self.data.X]))
# try with two clusters and less data
self.kmeans.set_data(self.data[:3])
self.kmeans.add_centroids([[4.7, 3.0]])
desired_array = np.array([
np.array([[5.100, 3.500]]),
np.array([[4.900, 3.000], [4.700, 3.200]])
])
for i, arr in enumerate(self.kmeans.centroids_belonging_points):
np.testing.assert_equal(arr, desired_array[i])
def test_step_completed(self):
centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans.add_centroids(centroids)
self.assertEqual(self.kmeans.step_completed, True)
self.kmeans.step()
self.assertEqual(self.kmeans.step_completed, False)
self.kmeans.step()
self.assertEqual(self.kmeans.step_completed, True)
def test_set_data(self):
self.kmeans.set_data(self.data)
self.assertEqual(self.kmeans.data, self.data)
self.assertEqual(self.kmeans.centroids_history, [])
self.assertEqual(self.kmeans.step_no, 0)
self.assertEqual(self.kmeans.step_completed, True)
self.assertEqual(self.kmeans.centroids_moved, False)
# try with none data
self.kmeans.set_data(None)
self.assertEqual(self.kmeans.data, None)
self.assertEqual(self.kmeans.centroids_history, [])
self.assertEqual(self.kmeans.clusters, None)
self.assertEqual(self.kmeans.step_no, 0)
self.assertEqual(self.kmeans.step_completed, True)
self.assertEqual(self.kmeans.centroids_moved, False)
def test_find_clusters(self):
self.kmeans.add_centroids([[5.2, 3.6]])
# if only one cluster all points in 0th element of first dimension
np.testing.assert_equal(
self.kmeans.find_clusters(self.kmeans.centroids),
np.zeros(len(self.data)))
# try with two clusters and less data
self.kmeans.set_data(self.data[:3])
self.kmeans.add_centroids([[4.7, 3.0]])
np.testing.assert_equal(
self.kmeans.find_clusters(self.kmeans.centroids),
np.array([0, 1, 1]))
def test_step(self):
centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans.add_centroids(centroids)
centroids_before = np.copy(self.kmeans.centroids)
clusters_before = np.copy(self.kmeans.clusters)
self.kmeans.step()
self.assertEqual(self.kmeans.step_completed, False)
self.assertEqual(self.kmeans.centroids_moved, True)
np.testing.assert_equal(centroids_before,
self.kmeans.centroids_history[-2])
# clusters doesnt change in every odd step
np.testing.assert_equal(clusters_before, self.kmeans.clusters)
centroids_before = np.copy(self.kmeans.centroids)
self.kmeans.step()
self.assertEqual(self.kmeans.step_completed, True)
self.assertEqual(self.kmeans.centroids_moved, False)
np.testing.assert_equal(centroids_before, self.kmeans.centroids)
centroids_before = np.copy(self.kmeans.centroids)
self.kmeans.step()
self.kmeans.step_back()
np.testing.assert_equal(centroids_before, self.kmeans.centroids)
def test_step_back(self):
centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans.add_centroids(centroids)
# check if nothing happens when step = 0
centroids_before = np.copy(self.kmeans.centroids)
clusters_before = np.copy(self.kmeans.clusters)
self.kmeans.step_back()
np.testing.assert_equal(centroids_before, self.kmeans.centroids)
np.testing.assert_equal(clusters_before, self.kmeans.clusters)
# check if centroids remain in even step
self.kmeans.step()
self.kmeans.step()
centroids_before = self.kmeans.centroids
self.kmeans.step_back()
np.testing.assert_equal(centroids_before, self.kmeans.centroids)
self.assertEqual(self.kmeans.step_completed, False)
self.assertEqual(self.kmeans.centroids_moved, False)
# check if clusters remain in even step
clusters_before = self.kmeans.clusters
self.kmeans.step_back()
np.testing.assert_equal(clusters_before, self.kmeans.clusters)
self.assertEqual(self.kmeans.step_completed, True)
self.assertEqual(self.kmeans.centroids_moved, True)
def test_random_positioning(self):
self.assertEqual(self.kmeans.random_positioning(4).shape, (4, 2))
self.assertEqual(self.kmeans.random_positioning(1).shape, (1, 2))
self.assertEqual(self.kmeans.random_positioning(0).shape, (0, ))
self.assertEqual(self.kmeans.random_positioning(-1).shape, (0, ))
def test_add_centroids(self):
self.kmeans.add_centroids([[5.2, 3.1]])
self.assertEqual(self.kmeans.k, 1)
self.kmeans.add_centroids([[6.5, 3], [7, 4]])
self.assertEqual(self.kmeans.k, 3)
self.kmeans.add_centroids(2)
self.assertEqual(self.kmeans.k, 5)
self.kmeans.add_centroids()
self.assertEqual(self.kmeans.k, 6)
step_before = self.kmeans.step_no
self.assertEqual(step_before, self.kmeans.step_no)
self.kmeans.step()
self.kmeans.add_centroids()
self.assertEqual(step_before + 2, self.kmeans.step_no)
self.assertEqual(self.kmeans.centroids_moved, False)
def test_delete_centroids(self):
self.kmeans.add_centroids([[6.5, 3], [7, 4], [5.2, 3.1]])
self.kmeans.delete_centroids(1)
self.assertEqual(self.kmeans.k, 2)
self.kmeans.delete_centroids(2)
self.assertEqual(self.kmeans.k, 0)
self.kmeans.delete_centroids(2)
self.assertEqual(self.kmeans.k, 0)
def test_move_centroid(self):
self.kmeans.add_centroids([[6.5, 3], [7, 4], [5.2, 3.1]])
self.kmeans.move_centroid(1, 3, 3.2)
np.testing.assert_equal(self.kmeans.centroids[1], np.array([3, 3.2]))
self.assertEqual(self.kmeans.k, 3)
self.kmeans.step()
self.kmeans.move_centroid(2, 3.2, 3.4)
self.assertEqual(self.kmeans.centroids_moved, False)
self.assertEqual(self.kmeans.step_no, 2)
def test_set_list(self):
# test adding Nones if list too short
self.assertEqual(self.kmeans.set_list([], 2, 1), [None, None, 1])
# test adding Nones if list too short
self.assertEqual(self.kmeans.set_list([2], 2, 1), [2, None, 1])
# adding to end
self.assertEqual(self.kmeans.set_list([2, 1], 2, 1), [2, 1, 1])
# changing the element in the last place
self.assertEqual(self.kmeans.set_list([2, 1], 1, 3), [2, 3])
# changing the element in the middle place
self.assertEqual(self.kmeans.set_list([2, 1, 3], 1, 3), [2, 3, 3])
class TestKmeans(unittest.TestCase):
def setUp(self):
self.data = Table('iris')
new_domain = Domain(self.data.domain.attributes[:2])
self.data = Table(new_domain, self.data)
# self.centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans = Kmeans(self.data, centroids=None)
def test_k(self):
centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans.add_centroids(centroids)
self.assertEqual(self.kmeans.k, 3)
self.assertEqual(self.kmeans.k, len(self.kmeans.centroids))
def test_centroids_belonging_points(self):
centroids = [[5.2, 3.6]]
self.kmeans.add_centroids(centroids)
# if only one cluster all points in 0th element of first dimension
np.testing.assert_equal(self.kmeans.centroids_belonging_points, np.array([self.data.X]))
# try with two clusters and less data
self.kmeans.set_data(self.data[:3])
self.kmeans.add_centroids([[4.7, 3.0]])
desired_array = np.array([np.array([[5.100, 3.500]]),
np.array([[4.900, 3.000],
[4.700, 3.200]])])
for i, arr in enumerate(self.kmeans.centroids_belonging_points):
np.testing.assert_equal(arr, desired_array[i])
def test_step_completed(self):
centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans.add_centroids(centroids)
self.assertEqual(self.kmeans.step_completed, True)
self.kmeans.step()
self.assertEqual(self.kmeans.step_completed, False)
self.kmeans.step()
self.assertEqual(self.kmeans.step_completed, True)
def test_set_data(self):
self.kmeans.set_data(self.data)
self.assertEqual(self.kmeans.data, self.data)
self.assertEqual(self.kmeans.centroids_history, [])
self.assertEqual(self.kmeans.stepNo, 0)
self.assertEqual(self.kmeans.step_completed, True)
self.assertEqual(self.kmeans.centroids_moved, False)
# try with none data
self.kmeans.set_data(None)
self.assertEqual(self.kmeans.data, None)
self.assertEqual(self.kmeans.centroids_history, [])
self.assertEqual(self.kmeans.clusters, None)
self.assertEqual(self.kmeans.stepNo, 0)
self.assertEqual(self.kmeans.step_completed, True)
self.assertEqual(self.kmeans.centroids_moved, False)
def test_find_clusters(self):
self.kmeans.add_centroids([[5.2, 3.6]])
# if only one cluster all points in 0th element of first dimension
np.testing.assert_equal(self.kmeans.find_clusters(self.kmeans.centroids), np.zeros(len(self.data)))
# try with two clusters and less data
self.kmeans.set_data(self.data[:3])
self.kmeans.add_centroids([[4.7, 3.0]])
np.testing.assert_equal(self.kmeans.find_clusters(self.kmeans.centroids), np.array([0, 1, 1]))
def test_step(self):
centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans.add_centroids(centroids)
centroids_before = np.copy(self.kmeans.centroids)
clusters_before = np.copy(self.kmeans.clusters)
self.kmeans.step()
self.assertEqual(self.kmeans.step_completed, False)
self.assertEqual(self.kmeans.centroids_moved, True)
np.testing.assert_equal(centroids_before, self.kmeans.centroids_history[-1])
# clusters doesnt change in every odd step
np.testing.assert_equal(clusters_before, self.kmeans.clusters)
centroids_before = np.copy(self.kmeans.centroids)
self.kmeans.step()
self.assertEqual(self.kmeans.step_completed, True)
self.assertEqual(self.kmeans.centroids_moved, False)
np.testing.assert_equal(centroids_before, self.kmeans.centroids)
centroids_before = np.copy(self.kmeans.centroids)
self.kmeans.step()
self.kmeans.step_back()
np.testing.assert_equal(centroids_before, self.kmeans.centroids)
def test_step_back(self):
centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans.add_centroids(centroids)
# check if nothing happens when step = 0
centroids_before = np.copy(self.kmeans.centroids)
clusters_before = np.copy(self.kmeans.clusters)
self.kmeans.step_back()
np.testing.assert_equal(centroids_before, self.kmeans.centroids)
np.testing.assert_equal(clusters_before, self.kmeans.clusters)
# check if centroids remain in even step
self.kmeans.step()
self.kmeans.step()
centroids_before = self.kmeans.centroids
self.kmeans.step_back()
np.testing.assert_equal(centroids_before, self.kmeans.centroids)
self.assertEqual(self.kmeans.step_completed, False)
self.assertEqual(self.kmeans.centroids_moved, False)
# check if clusters remain in even step
clusters_before = self.kmeans.clusters
self.kmeans.step_back()
np.testing.assert_equal(clusters_before, self.kmeans.clusters)
self.assertEqual(self.kmeans.step_completed, True)
self.assertEqual(self.kmeans.centroids_moved, True)
def test_random_positioning(self):
self.assertEqual(self.kmeans.random_positioning(4).shape, (4, 2))
self.assertEqual(self.kmeans.random_positioning(1).shape, (1, 2))
self.assertEqual(self.kmeans.random_positioning(0).shape, (0,))
self.assertEqual(self.kmeans.random_positioning(-1).shape, (0,))
def test_add_centroids(self):
self.kmeans.add_centroids([[5.2, 3.1]])
self.assertEqual(self.kmeans.k, 1)
self.kmeans.add_centroids([[6.5, 3], [7, 4]])
self.assertEqual(self.kmeans.k, 3)
self.kmeans.add_centroids(2)
self.assertEqual(self.kmeans.k, 5)
self.kmeans.add_centroids()
self.assertEqual(self.kmeans.k, 6)
step_before = self.kmeans.stepNo
self.assertEqual(step_before, self.kmeans.stepNo)
self.kmeans.step()
self.kmeans.add_centroids()
self.assertEqual(step_before + 2, self.kmeans.stepNo)
self.assertEqual(self.kmeans.centroids_moved, False)
def test_delete_centroids(self):
self.kmeans.add_centroids([[6.5, 3], [7, 4], [5.2, 3.1]])
self.kmeans.delete_centroids(1)
self.assertEqual(self.kmeans.k, 2)
self.kmeans.delete_centroids(2)
self.assertEqual(self.kmeans.k, 0)
self.kmeans.delete_centroids(2)
self.assertEqual(self.kmeans.k, 0)
def test_move_centroid(self):
self.kmeans.add_centroids([[6.5, 3], [7, 4], [5.2, 3.1]])
self.kmeans.move_centroid(1, 3, 3.2)
np.testing.assert_equal(self.kmeans.centroids[1], np.array([3, 3.2]))
self.assertEqual(self.kmeans.k, 3)
self.kmeans.step()
self.kmeans.move_centroid(2, 3.2, 3.4)
self.assertEqual(self.kmeans.centroids_moved, False)
self.assertEqual(self.kmeans.stepNo, 2)
class OWKmeans(OWWidget):
"""
K-means widget
"""
name = "Interactive k-Means"
description = "Widget demonstrates working of k-means algorithm."
keywords = ["kmeans", "clustering", "interactive"]
icon = "icons/InteractiveKMeans.svg"
want_main_area = False
priority = 300
# inputs and outputs
class Inputs:
data = Input("Data", Table)
class Outputs:
annotated_data = Output("Annotated Data", Table, default=True)
centroids = Output("Centroids", Table)
class Warning(OWWidget.Warning):
num_features = Msg(
"Widget requires at least two numeric features with valid values")
cluster_points = Msg(
"The number of clusters can't exceed the number of points")
# settings
number_of_clusters = settings.Setting(3)
auto_play_enabled = False
auto_play_thread = None
# data
data = None
selected_rows = None # rows that are selected for kmeans (not nan rows)
# selected attributes in chart
attr_x = settings.Setting('')
attr_y = settings.Setting('')
# other settings
k_means = None
auto_play_speed = settings.Setting(1)
lines_to_centroids = settings.Setting(True)
graph_name = 'scatter'
output_name = "cluster"
STEP_BUTTONS = ["Reassign Membership", "Recompute Centroids"]
AUTOPLAY_BUTTONS = ["Run", "Stop"]
# colors taken from chart.options.colors in Highchart
# (if more required check for more in chart.options.color)
colors = [
"#1F7ECA", "#D32525", "#28D825", "#D5861F", "#98257E", "#2227D5",
"#D5D623", "#D31BD6", "#6A7CDB", "#78D5D4"
]
# signals
step_trigger = pyqtSignal()
stop_auto_play_trigger = pyqtSignal()
def __init__(self):
super().__init__()
# options box
self.options_box = gui.widgetBox(self.controlArea, "Data")
opts = dict(widget=self.options_box,
master=self,
orientation=Qt.Horizontal,
callback=self.restart,
sendSelectedValue=True,
maximumContentsLength=15)
self.cbx = gui.comboBox(value='attr_x', label='X: ', **opts)
self.cby = gui.comboBox(value='attr_y', label='Y: ', **opts)
self.centroids_box = gui.widgetBox(self.controlArea, "Centroids")
self.centroid_numbers_spinner = gui.spin(
self.centroids_box,
self,
'number_of_clusters',
minv=1,
maxv=10,
step=1,
label='Number of centroids:',
alignment=Qt.AlignRight,
callback=self.number_of_clusters_change)
self.restart_button = gui.button(self.centroids_box,
self,
"Randomize Positions",
callback=self.restart)
gui.separator(self.centroids_box)
self.lines_checkbox = gui.checkBox(self.centroids_box,
self,
'lines_to_centroids',
'Show membership lines',
callback=self.complete_replot)
# control box
gui.separator(self.controlArea, 20, 20)
self.step_box = gui.widgetBox(self.controlArea,
"Manually step through")
self.step_button = gui.button(self.step_box,
self,
self.STEP_BUTTONS[1],
callback=self.step)
self.step_back_button = gui.button(self.step_box,
self,
"Step Back",
callback=self.step_back)
self.run_box = gui.widgetBox(self.controlArea, "Run")
self.auto_play_speed_spinner = gui.hSlider(self.run_box,
self,
'auto_play_speed',
label='Speed:',
minValue=0,
maxValue=1.91,
step=0.1,
intOnly=False,
createLabel=False)
self.auto_play_button = gui.button(self.run_box,
self,
self.AUTOPLAY_BUTTONS[0],
callback=self.auto_play)
gui.rubber(self.controlArea)
# disable until data loaded
self.set_disabled_all(True)
# graph in mainArea
self.scatter = Scatterplot(click_callback=self.graph_clicked,
drop_callback=self.centroid_dropped,
xAxis_gridLineWidth=0,
yAxis_gridLineWidth=0,
tooltip_enabled=False,
debug=False)
# Just render an empty chart so it shows a nice 'No data to display'
self.scatter.chart()
self.mainArea.layout().addWidget(self.scatter)
def concat_x_y(self):
"""
Function takes two selected columns from data table and merge them in
new Orange.data.Table
Returns
-------
Orange.data.Table
table with selected columns
"""
attr_x = self.data.domain[self.attr_x]
attr_y = self.data.domain[self.attr_y]
cols = []
for attr in (attr_x, attr_y):
subset = self.data[:, attr]
cols.append(subset.Y if subset.Y.size else subset.X)
x = np.column_stack(cols)
not_nan = ~np.isnan(x).any(axis=1)
x = x[not_nan] # remove rows with nan
self.selected_rows = np.where(not_nan)
domain = Domain([attr_x, attr_y])
return Table(domain, x)
def set_empty_plot(self):
self.scatter.clear()
def set_disabled_all(self, disabled):
"""
Function disable all controls
"""
self.options_box.setDisabled(disabled)
self.centroids_box.setDisabled(disabled)
self.step_box.setDisabled(disabled)
self.run_box.setDisabled(disabled)
@Inputs.data
def set_data(self, data):
"""
Function receives data from input and init part of widget if data are
ok. Otherwise set empty plot and notice
user about that
Parameters
----------
data : Orange.data.Table or None
input data
"""
self.data = data
def get_valid_attributes(data):
attrs = [
var for var in data.domain.attributes if var.is_continuous
]
return [var for var in attrs if sum(~np.isnan(data[:, var])) > 0]
def reset_combos():
self.cbx.clear()
self.cby.clear()
def init_combos():
"""
function initialize the combos with attributes
"""
reset_combos()
valid_class_vars = [
var for var in data.domain.class_vars
if data is not None and var.is_continuous
]
for var in chain(valid_attributes, valid_class_vars):
self.cbx.addItem(gui.attributeIconDict[var], var.name)
self.cby.addItem(gui.attributeIconDict[var], var.name)
# remove warnings about too less continuous attributes and not enough data
self.Warning.clear()
if self.auto_play_thread:
self.auto_play_thread.stop()
if data is None or len(data) == 0:
reset_combos()
self.set_empty_plot()
self.set_disabled_all(True)
return
valid_attributes = get_valid_attributes(data)
if len(valid_attributes) < 2:
reset_combos()
self.Warning.num_features()
self.set_empty_plot()
self.set_disabled_all(True)
else:
init_combos()
self.set_disabled_all(False)
self.attr_x = self.cbx.itemText(0)
self.attr_y = self.cbx.itemText(1)
if self.k_means is None:
self.k_means = Kmeans(self.concat_x_y())
else:
self.k_means.set_data(self.concat_x_y())
self.number_of_clusters_change()
def restart(self):
"""
Function triggered on data change or restart button pressed
"""
self.k_means = Kmeans(self.concat_x_y())
self.number_of_clusters_change()
def step(self):
"""
Function called on every step
"""
self.k_means.step()
self.replot()
self.button_text_change()
self.send_data()
def step_back(self):
"""
Function called for step back
"""
self.k_means.step_back()
self.replot()
self.button_text_change()
self.send_data()
self.number_of_clusters = self.k_means.k
def button_text_change(self):
"""
Function changes text on ste button and chanbe the button text
"""
self.step_button.setText(
self.STEP_BUTTONS[self.k_means.step_completed])
if self.k_means.step_no <= 0:
self.step_back_button.setDisabled(True)
elif not self.auto_play_enabled:
self.step_back_button.setDisabled(False)
def auto_play(self):
"""
Function called when autoplay button pressed
"""
self.auto_play_enabled = not self.auto_play_enabled
self.auto_play_button.setText(
self.AUTOPLAY_BUTTONS[self.auto_play_enabled])
if self.auto_play_enabled:
self.options_box.setDisabled(True)
self.centroids_box.setDisabled(True)
self.step_box.setDisabled(True)
self.auto_play_thread = Autoplay(self)
self.step_trigger.connect(self.step)
self.stop_auto_play_trigger.connect(self.stop_auto_play)
self.auto_play_thread.start()
else:
self.stop_auto_play()
def stop_auto_play(self):
"""
Called when stop autoplay button pressed or in the end of autoplay
"""
self.options_box.setDisabled(False)
self.centroids_box.setDisabled(False)
self.step_box.setDisabled(False)
self.auto_play_enabled = False
self.auto_play_button\
.setText(self.AUTOPLAY_BUTTONS[self.auto_play_enabled])
self.button_text_change()
def replot(self):
"""
Function refreshes the chart
"""
if self.data is None or not self.attr_x or not self.attr_y:
return
km = self.k_means
if not km.centroids_moved:
self.complete_replot()
return
# when centroids moved during step
self.scatter.update_series(0, self.k_means.centroids)
if self.lines_to_centroids:
for i, (c, pts) in enumerate(
zip(km.centroids, km.centroids_belonging_points)):
self.scatter.update_series(
1 + i,
list(
chain.from_iterable(
([p[0], p[1]], [c[0], c[1]]) for p in pts)))
def complete_replot(self):
"""
This function performs complete replot of the graph without animation
"""
try:
attr_x = self.data.domain[self.attr_x]
attr_y = self.data.domain[self.attr_y]
except KeyError:
return
# plot centroids
options = dict(series=[])
n_colors = len(self.colors)
km = self.k_means
options['series'].append(
dict(data=[{
'x': p[0],
'y': p[1],
'marker': {
'fillColor': self.colors[i % n_colors]
}
} for i, p in enumerate(km.centroids)],
type="scatter",
draggableX=True,
draggableY=True,
cursor="move",
zIndex=10,
marker=dict(symbol='square', radius=8)))
# plot lines between centroids and points
if self.lines_to_centroids:
for i, (c, pts) in enumerate(
zip(km.centroids, km.centroids_belonging_points)):
options['series'].append(
dict(data=list(
chain.from_iterable(
([p[0], p[1]], [c[0], c[1]]) for p in pts)),
type="line",
lineWidth=0.2,
enableMouseTracking=False,
color="#ccc"))
# plot data points
for i, points in enumerate(km.centroids_belonging_points):
options['series'].append(
dict(data=points,
type="scatter",
color=rgb_hash_brighter(self.colors[i % len(self.colors)],
0.3)))
# highcharts parameters
kwargs = dict(
xAxis_title_text=attr_x.name,
yAxis_title_text=attr_y.name,
tooltip_headerFormat="",
tooltip_pointFormat="<strong>%s:</strong> {point.x:.2f} <br/>"
"<strong>%s:</strong> {point.y:.2f}" % (self.attr_x, self.attr_y))
# plot
self.scatter.chart(options, **kwargs)
def replot_series(self):
"""
This function replot just series connected with centroids and
uses animation for that
"""
km = self.k_means
k = km.k
series = []
# plot lines between centroids and points
if self.lines_to_centroids:
for i, (c, pts) in enumerate(
zip(km.centroids, km.centroids_belonging_points)):
series.append(
dict(data=list(
chain.from_iterable(
([p[0], p[1]], [c[0], c[1]]) for p in pts)),
type="line",
showInLegend=False,
lineWidth=0.2,
enableMouseTracking=False,
color="#ccc"))
# plot data points
for i, points in enumerate(km.centroids_belonging_points):
series.append(
dict(data=points,
type="scatter",
showInLegend=False,
color=rgb_hash_brighter(self.colors[i % len(self.colors)],
0.5)))
self.scatter.add_series(series)
self.scatter.remove_last_series(k *
2 if self.lines_to_centroids else k)
def number_of_clusters_change(self):
"""
Function that change number of clusters if required
"""
if self.data is None:
return
if self.number_of_clusters > len(self.data):
# if too less data for clusters number
self.Warning.cluster_points()
self.set_empty_plot()
self.step_box.setDisabled(True)
self.run_box.setDisabled(True)
else:
self.Warning.cluster_points.clear()
self.step_box.setDisabled(False)
self.run_box.setDisabled(False)
if self.k_means is None: # if before too less data k_means is None
self.k_means = Kmeans(self.concat_x_y())
if self.k_means.k < self.number_of_clusters:
self.k_means.add_centroids(self.number_of_clusters -
self.k_means.k)
elif not self.k_means.k == self.number_of_clusters:
self.k_means.delete_centroids(self.k_means.k -
self.number_of_clusters)
self.replot()
self.send_data()
self.button_text_change()
def graph_clicked(self, x, y):
"""
Function called when user click in graph. Centroid have to be added.
"""
if self.k_means is not None and self.data is not None:
self.k_means.add_centroids([x, y])
self.number_of_clusters += 1
self.replot()
self.send_data()
self.button_text_change()
def centroid_dropped(self, _index, x, y):
"""
Function called when centroid with _index moved.
"""
self.k_means.move_centroid(_index, x, y)
self.complete_replot()
self.send_data()
self.button_text_change()
def send_data(self):
"""
Function sends data with clusters column and data with centroids
position to the output
"""
km = self.k_means
if km is None or km.clusters is None:
self.Outputs.annotated_data.send(None)
self.Outputs.centroids.send(None)
else:
clust_var = DiscreteVariable(
self.output_name,
values=["C%d" % (x + 1) for x in range(km.k)])
attributes = self.data.domain.attributes
classes = self.data.domain.class_vars
meta_attrs = self.data.domain.metas
if classes:
meta_attrs += classes
classes = [clust_var]
domain = Domain(attributes, classes, meta_attrs)
annotated_data = Table.from_table(domain, self.data)
annotated_data.Y[self.selected_rows] = km.clusters
centroids = Table(Domain(km.data.domain.attributes), km.centroids)
self.Outputs.annotated_data.send(annotated_data)
self.Outputs.centroids.send(centroids)
def send_report(self):
if self.data is None:
return
caption = report.render_items_vert(
(("Number of centroids:", self.number_of_clusters), ))
self.report_plot(self.scatter)
self.report_caption(caption)
class TestKmeans(unittest.TestCase):
def setUp(self):
self.data = Table('iris')
new_domain = Domain(self.data.domain.attributes[:2])
self.data = Table(new_domain, self.data)
# self.centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans = Kmeans(self.data)
def test_k(self):
centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans.add_centroids(centroids)
self.assertEqual(self.kmeans.k, 3)
self.assertEqual(self.kmeans.k, len(self.kmeans.centroids))
def test_converged(self):
centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans.add_centroids(centroids)
self.assertFalse(self.kmeans.converged)
self.kmeans.step()
self.assertFalse(self.kmeans.converged)
# step not complete so false every odd state
self.kmeans.step()
# it is even step so maybe converged but it depends on example
# unable to test
self.kmeans.step()
self.assertFalse(self.kmeans.converged)
# check if false every not completed step
for i in range(self.kmeans.max_iter // 2 + 1):
self.kmeans.step()
self.kmeans.step()
self.assertFalse(self.kmeans.converged)
# it converged because of max iter
self.kmeans.step()
self.assertTrue(self.kmeans.converged)
def test_centroids_belonging_points(self):
centroids = [[5.2, 3.6]]
self.kmeans.add_centroids(centroids)
# if only one cluster all points in 0th element of first dimension
np.testing.assert_equal(
self.kmeans.centroids_belonging_points, np.array([self.data.X]))
# try with two clusters and less data
self.kmeans.set_data(self.data[:3])
self.kmeans.add_centroids([[4.7, 3.0]])
desired_array = np.array([np.array([[5.100, 3.500]]),
np.array([[4.900, 3.000],
[4.700, 3.200]])])
for i, arr in enumerate(self.kmeans.centroids_belonging_points):
np.testing.assert_equal(arr, desired_array[i])
def test_step_completed(self):
centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans.add_centroids(centroids)
self.assertEqual(self.kmeans.step_completed, True)
self.kmeans.step()
self.assertEqual(self.kmeans.step_completed, False)
self.kmeans.step()
self.assertEqual(self.kmeans.step_completed, True)
def test_set_data(self):
self.kmeans.set_data(self.data)
self.assertEqual(self.kmeans.data, self.data)
self.assertEqual(self.kmeans.centroids_history, [])
self.assertEqual(self.kmeans.step_no, 0)
self.assertEqual(self.kmeans.step_completed, True)
self.assertEqual(self.kmeans.centroids_moved, False)
# try with none data
self.kmeans.set_data(None)
self.assertEqual(self.kmeans.data, None)
self.assertEqual(self.kmeans.centroids_history, [])
self.assertEqual(self.kmeans.clusters, None)
self.assertEqual(self.kmeans.step_no, 0)
self.assertEqual(self.kmeans.step_completed, True)
self.assertEqual(self.kmeans.centroids_moved, False)
def test_find_clusters(self):
self.kmeans.add_centroids([[5.2, 3.6]])
# if only one cluster all points in 0th element of first dimension
np.testing.assert_equal(
self.kmeans.find_clusters(self.kmeans.centroids),
np.zeros(len(self.data)))
# try with two clusters and less data
self.kmeans.set_data(self.data[:3])
self.kmeans.add_centroids([[4.7, 3.0]])
np.testing.assert_equal(
self.kmeans.find_clusters(self.kmeans.centroids),
np.array([0, 1, 1]))
def test_step(self):
centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans.add_centroids(centroids)
centroids_before = np.copy(self.kmeans.centroids)
clusters_before = np.copy(self.kmeans.clusters)
self.kmeans.step()
self.assertEqual(self.kmeans.step_completed, False)
self.assertEqual(self.kmeans.centroids_moved, True)
np.testing.assert_equal(centroids_before,
self.kmeans.centroids_history[-2])
# clusters doesnt change in every odd step
np.testing.assert_equal(clusters_before, self.kmeans.clusters)
centroids_before = np.copy(self.kmeans.centroids)
self.kmeans.step()
self.assertEqual(self.kmeans.step_completed, True)
self.assertEqual(self.kmeans.centroids_moved, False)
np.testing.assert_equal(centroids_before, self.kmeans.centroids)
centroids_before = np.copy(self.kmeans.centroids)
self.kmeans.step()
self.kmeans.step_back()
np.testing.assert_equal(centroids_before, self.kmeans.centroids)
def test_step_back(self):
centroids = [[5.2, 3.1], [6.5, 3], [7, 4]]
self.kmeans.add_centroids(centroids)
# check if nothing happens when step = 0
centroids_before = np.copy(self.kmeans.centroids)
clusters_before = np.copy(self.kmeans.clusters)
self.kmeans.step_back()
np.testing.assert_equal(centroids_before, self.kmeans.centroids)
np.testing.assert_equal(clusters_before, self.kmeans.clusters)
# check if centroids remain in even step
self.kmeans.step()
self.kmeans.step()
centroids_before = self.kmeans.centroids
self.kmeans.step_back()
np.testing.assert_equal(centroids_before, self.kmeans.centroids)
self.assertEqual(self.kmeans.step_completed, False)
self.assertEqual(self.kmeans.centroids_moved, False)
# check if clusters remain in even step
clusters_before = self.kmeans.clusters
self.kmeans.step_back()
np.testing.assert_equal(clusters_before, self.kmeans.clusters)
self.assertEqual(self.kmeans.step_completed, True)
self.assertEqual(self.kmeans.centroids_moved, True)
def test_random_positioning(self):
self.assertEqual(self.kmeans.random_positioning(4).shape, (4, 2))
self.assertEqual(self.kmeans.random_positioning(1).shape, (1, 2))
self.assertEqual(self.kmeans.random_positioning(0).shape, (0,))
self.assertEqual(self.kmeans.random_positioning(-1).shape, (0,))
def test_add_centroids(self):
self.kmeans.add_centroids([[5.2, 3.1]])
self.assertEqual(self.kmeans.k, 1)
self.kmeans.add_centroids([[6.5, 3], [7, 4]])
self.assertEqual(self.kmeans.k, 3)
self.kmeans.add_centroids(2)
self.assertEqual(self.kmeans.k, 5)
self.kmeans.add_centroids()
self.assertEqual(self.kmeans.k, 6)
step_before = self.kmeans.step_no
self.assertEqual(step_before, self.kmeans.step_no)
self.kmeans.step()
self.kmeans.add_centroids()
self.assertEqual(step_before + 2, self.kmeans.step_no)
self.assertEqual(self.kmeans.centroids_moved, False)
def test_delete_centroids(self):
self.kmeans.add_centroids([[6.5, 3], [7, 4], [5.2, 3.1]])
self.kmeans.delete_centroids(1)
self.assertEqual(self.kmeans.k, 2)
self.kmeans.delete_centroids(2)
self.assertEqual(self.kmeans.k, 0)
self.kmeans.delete_centroids(2)
self.assertEqual(self.kmeans.k, 0)
def test_move_centroid(self):
self.kmeans.add_centroids([[6.5, 3], [7, 4], [5.2, 3.1]])
self.kmeans.move_centroid(1, 3, 3.2)
np.testing.assert_equal(self.kmeans.centroids[1], np.array([3, 3.2]))
self.assertEqual(self.kmeans.k, 3)
self.kmeans.step()
self.kmeans.move_centroid(2, 3.2, 3.4)
self.assertEqual(self.kmeans.centroids_moved, False)
self.assertEqual(self.kmeans.step_no, 2)
def test_set_list(self):
# test adding Nones if list too short
self.assertEqual(self.kmeans.set_list([], 2, 1), [None, None, 1])
# test adding Nones if list too short
self.assertEqual(self.kmeans.set_list([2], 2, 1), [2, None, 1])
# adding to end
self.assertEqual(self.kmeans.set_list([2, 1], 2, 1), [2, 1, 1])
# changing the element in the last place
self.assertEqual(self.kmeans.set_list([2, 1], 1, 3), [2, 3])
# changing the element in the middle place
self.assertEqual(self.kmeans.set_list([2, 1, 3], 1, 3), [2, 3, 3])
class OWKmeans(OWWidget):
"""
K-means widget
"""
name = "Interactive k-Means"
description = "Widget demonstrates working of k-means algorithm."
icon = "icons/InteractiveKMeans.svg"
want_main_area = False
# inputs and outputs
class Inputs:
data = Input("Data", Table)
class Outputs:
annotated_data = Output("Annotated Data", Table, default=True)
centroids = Output("Centroids", Table)
class Warning(OWWidget.Warning):
num_features = Msg("Widget requires at least two numeric features with valid values")
cluster_points = Msg("The number of clusters can't exceed the number of points")
# settings
number_of_clusters = settings.Setting(3)
auto_play_enabled = False
auto_play_thread = None
# data
data = None
selected_rows = None # rows that are selected for kmeans (not nan rows)
# selected attributes in chart
attr_x = settings.Setting('')
attr_y = settings.Setting('')
# other settings
k_means = None
auto_play_speed = settings.Setting(1)
lines_to_centroids = settings.Setting(True)
graph_name = 'scatter'
output_name = "cluster"
STEP_BUTTONS = ["Reassign Membership", "Recompute Centroids"]
AUTOPLAY_BUTTONS = ["Run", "Stop"]
# colors taken from chart.options.colors in Highchart
# (if more required check for more in chart.options.color)
colors = ["#1F7ECA", "#D32525", "#28D825", "#D5861F", "#98257E",
"#2227D5", "#D5D623", "#D31BD6", "#6A7CDB", "#78D5D4"]
# signals
step_trigger = pyqtSignal()
stop_auto_play_trigger = pyqtSignal()
def __init__(self):
super().__init__()
# options box
self.options_box = gui.widgetBox(self.controlArea, "Data")
opts = dict(
widget=self.options_box, master=self, orientation=Qt.Horizontal,
callback=self.restart, sendSelectedValue=True,
maximumContentsLength=15)
self.cbx = gui.comboBox(value='attr_x', label='X: ', **opts)
self.cby = gui.comboBox(value='attr_y', label='Y: ', **opts)
self.centroids_box = gui.widgetBox(self.controlArea, "Centroids")
self.centroid_numbers_spinner = gui.spin(
self.centroids_box, self, 'number_of_clusters',
minv=1, maxv=10, step=1, label='Number of centroids:',
alignment=Qt.AlignRight, callback=self.number_of_clusters_change)
self.restart_button = gui.button(
self.centroids_box, self, "Randomize Positions",
callback=self.restart)
gui.separator(self.centroids_box)
self.lines_checkbox = gui.checkBox(
self.centroids_box, self, 'lines_to_centroids',
'Show membership lines', callback=self.complete_replot)
# control box
gui.separator(self.controlArea, 20, 20)
self.step_box = gui.widgetBox(self.controlArea, "Manually step through")
self.step_button = gui.button(
self.step_box, self, self.STEP_BUTTONS[1], callback=self.step)
self.step_back_button = gui.button(
self.step_box, self, "Step Back", callback=self.step_back)
self.run_box = gui.widgetBox(self.controlArea, "Run")
self.auto_play_speed_spinner = gui.hSlider(
self.run_box, self, 'auto_play_speed', label='Speed:',
minValue=0, maxValue=1.91, step=0.1, intOnly=False,
createLabel=False)
self.auto_play_button = gui.button(
self.run_box, self, self.AUTOPLAY_BUTTONS[0],
callback=self.auto_play)
gui.rubber(self.controlArea)
# disable until data loaded
self.set_disabled_all(True)
# graph in mainArea
self.scatter = Scatterplot(
click_callback=self.graph_clicked,
drop_callback=self.centroid_dropped,
xAxis_gridLineWidth=0, yAxis_gridLineWidth=0,
tooltip_enabled=False,
debug=False)
# Just render an empty chart so it shows a nice 'No data to display'
self.scatter.chart()
self.mainArea.layout().addWidget(self.scatter)
def concat_x_y(self):
"""
Function takes two selected columns from data table and merge them in
new Orange.data.Table
Returns
-------
Orange.data.Table
table with selected columns
"""
attr_x = self.data.domain[self.attr_x]
attr_y = self.data.domain[self.attr_y]
cols = []
for attr in (attr_x, attr_y):
subset = self.data[:, attr]
cols.append(subset.Y if subset.Y.size else subset.X)
x = np.column_stack(cols)
not_nan = ~np.isnan(x).any(axis=1)
x = x[not_nan] # remove rows with nan
self.selected_rows = np.where(not_nan)
domain = Domain([attr_x, attr_y])
return Table(domain, x)
def set_empty_plot(self):
self.scatter.clear()
def set_disabled_all(self, disabled):
"""
Function disable all controls
"""
self.options_box.setDisabled(disabled)
self.centroids_box.setDisabled(disabled)
self.step_box.setDisabled(disabled)
self.run_box.setDisabled(disabled)
@Inputs.data
def set_data(self, data):
"""
Function receives data from input and init part of widget if data are
ok. Otherwise set empty plot and notice
user about that
Parameters
----------
data : Orange.data.Table or None
input data
"""
self.data = data
def get_valid_attributes(data):
attrs = [var for var in data.domain.attributes if var.is_continuous]
return [var for var in attrs if sum(~np.isnan(data[:, var])) > 0]
def reset_combos():
self.cbx.clear()
self.cby.clear()
def init_combos():
"""
function initialize the combos with attributes
"""
reset_combos()
valid_class_vars = [var for var in data.domain.class_vars
if data is not None and var.is_continuous]
for var in chain(valid_attributes, valid_class_vars):
self.cbx.addItem(gui.attributeIconDict[var], var.name)
self.cby.addItem(gui.attributeIconDict[var], var.name)
# remove warnings about too less continuous attributes and not enough data
self.Warning.clear()
if self.auto_play_thread:
self.auto_play_thread.stop()
if data is None or len(data) == 0:
reset_combos()
self.set_empty_plot()
self.set_disabled_all(True)
return
valid_attributes = get_valid_attributes(data)
if len(valid_attributes) < 2:
reset_combos()
self.Warning.num_features()
self.set_empty_plot()
self.set_disabled_all(True)
else:
init_combos()
self.set_disabled_all(False)
self.attr_x = self.cbx.itemText(0)
self.attr_y = self.cbx.itemText(1)
if self.k_means is None:
self.k_means = Kmeans(self.concat_x_y())
else:
self.k_means.set_data(self.concat_x_y())
self.number_of_clusters_change()
def restart(self):
"""
Function triggered on data change or restart button pressed
"""
self.k_means = Kmeans(self.concat_x_y())
self.number_of_clusters_change()
def step(self):
"""
Function called on every step
"""
self.k_means.step()
self.replot()
self.button_text_change()
self.send_data()
def step_back(self):
"""
Function called for step back
"""
self.k_means.step_back()
self.replot()
self.button_text_change()
self.send_data()
self.number_of_clusters = self.k_means.k
def button_text_change(self):
"""
Function changes text on ste button and chanbe the button text
"""
self.step_button.setText(self.STEP_BUTTONS[self.k_means.step_completed])
if self.k_means.step_no <= 0:
self.step_back_button.setDisabled(True)
elif not self.auto_play_enabled:
self.step_back_button.setDisabled(False)
def auto_play(self):
"""
Function called when autoplay button pressed
"""
self.auto_play_enabled = not self.auto_play_enabled
self.auto_play_button.setText(
self.AUTOPLAY_BUTTONS[self.auto_play_enabled])
if self.auto_play_enabled:
self.options_box.setDisabled(True)
self.centroids_box.setDisabled(True)
self.step_box.setDisabled(True)
self.auto_play_thread = Autoplay(self)
self.step_trigger.connect(self.step)
self.stop_auto_play_trigger.connect(self.stop_auto_play)
self.auto_play_thread.start()
else:
self.stop_auto_play()
def stop_auto_play(self):
"""
Called when stop autoplay button pressed or in the end of autoplay
"""
self.options_box.setDisabled(False)
self.centroids_box.setDisabled(False)
self.step_box.setDisabled(False)
self.auto_play_enabled = False
self.auto_play_button\
.setText(self.AUTOPLAY_BUTTONS[self.auto_play_enabled])
self.button_text_change()
def replot(self):
"""
Function refreshes the chart
"""
if self.data is None or not self.attr_x or not self.attr_y:
return
km = self.k_means
if not km.centroids_moved:
self.complete_replot()
return
# when centroids moved during step
self.scatter.update_series(0, self.k_means.centroids)
if self.lines_to_centroids:
for i, (c, pts) in enumerate(zip(
km.centroids, km.centroids_belonging_points)):
self.scatter.update_series(1 + i, list(chain.from_iterable(
([p[0], p[1]], [c[0], c[1]])
for p in pts)))
def complete_replot(self):
"""
This function performs complete replot of the graph without animation
"""
try:
attr_x = self.data.domain[self.attr_x]
attr_y = self.data.domain[self.attr_y]
except KeyError:
return
# plot centroids
options = dict(series=[])
n_colors = len(self.colors)
km = self.k_means
options['series'].append(
dict(
data=[{'x': p[0], 'y': p[1],
'marker':{'fillColor': self.colors[i % n_colors]}}
for i, p in enumerate(km.centroids)],
type="scatter",
draggableX=True,
draggableY=True,
cursor="move",
zIndex=10,
marker=dict(symbol='square', radius=8)))
# plot lines between centroids and points
if self.lines_to_centroids:
for i, (c, pts) in enumerate(zip(
km.centroids, km.centroids_belonging_points)):
options['series'].append(dict(
data=list(
chain.from_iterable(([p[0], p[1]], [c[0], c[1]])
for p in pts)),
type="line",
lineWidth=0.2,
enableMouseTracking=False,
color="#ccc"))
# plot data points
for i, points in enumerate(km.centroids_belonging_points):
options['series'].append(dict(
data=points,
type="scatter",
color=rgb_hash_brighter(
self.colors[i % len(self.colors)], 0.3)))
# highcharts parameters
kwargs = dict(
xAxis_title_text=attr_x.name,
yAxis_title_text=attr_y.name,
tooltip_headerFormat="",
tooltip_pointFormat="<strong>%s:</strong> {point.x:.2f} <br/>"
"<strong>%s:</strong> {point.y:.2f}" %
(self.attr_x, self.attr_y))
# plot
self.scatter.chart(options, **kwargs)
def replot_series(self):
"""
This function replot just series connected with centroids and
uses animation for that
"""
km = self.k_means
k = km.k
series = []
# plot lines between centroids and points
if self.lines_to_centroids:
for i, (c, pts) in enumerate(zip(
km.centroids, km.centroids_belonging_points)):
series.append(dict(
data=list(
chain.from_iterable(([p[0], p[1]], [c[0], c[1]])
for p in pts)),
type="line",
showInLegend=False,
lineWidth=0.2,
enableMouseTracking=False,
color="#ccc"))
# plot data points
for i, points in enumerate(km.centroids_belonging_points):
series.append(dict(
data=points,
type="scatter",
showInLegend=False,
color=rgb_hash_brighter(
self.colors[i % len(self.colors)], 0.5)))
self.scatter.add_series(series)
self.scatter.remove_last_series(k * 2 if self.lines_to_centroids else k)
def number_of_clusters_change(self):
"""
Function that change number of clusters if required
"""
if self.data is None:
return
if self.number_of_clusters > len(self.data):
# if too less data for clusters number
self.Warning.cluster_points()
self.set_empty_plot()
self.step_box.setDisabled(True)
self.run_box.setDisabled(True)
else:
self.Warning.cluster_points.clear()
self.step_box.setDisabled(False)
self.run_box.setDisabled(False)
if self.k_means is None: # if before too less data k_means is None
self.k_means = Kmeans(self.concat_x_y())
if self.k_means.k < self.number_of_clusters:
self.k_means.add_centroids(
self.number_of_clusters - self.k_means.k)
elif not self.k_means.k == self.number_of_clusters:
self.k_means.delete_centroids(
self.k_means.k - self.number_of_clusters)
self.replot()
self.send_data()
self.button_text_change()
def graph_clicked(self, x, y):
"""
Function called when user click in graph. Centroid have to be added.
"""
if self.k_means is not None and self.data is not None:
self.k_means.add_centroids([x, y])
self.number_of_clusters += 1
self.replot()
self.send_data()
self.button_text_change()
def centroid_dropped(self, _index, x, y):
"""
Function called when centroid with _index moved.
"""
self.k_means.move_centroid(_index, x, y)
self.complete_replot()
self.send_data()
self.button_text_change()
def send_data(self):
"""
Function sends data with clusters column and data with centroids
position to the output
"""
km = self.k_means
if km is None or km.clusters is None:
self.Outputs.annotated_data.send(None)
self.Outputs.centroids.send(None)
else:
clust_var = DiscreteVariable(
self.output_name,
values=["C%d" % (x + 1) for x in range(km.k)])
attributes = self.data.domain.attributes
classes = self.data.domain.class_vars
meta_attrs = self.data.domain.metas
if classes:
meta_attrs += classes
classes = [clust_var]
domain = Domain(attributes, classes, meta_attrs)
annotated_data = Table.from_table(domain, self.data)
annotated_data.Y[self.selected_rows] = km.clusters
centroids = Table(Domain(km.data.domain.attributes), km.centroids)
self.Outputs.annotated_data.send(annotated_data)
self.Outputs.centroids.send(centroids)
def send_report(self):
if self.data is None:
return
caption = report.render_items_vert((
("Number of centroids:", self.number_of_clusters),
))
self.report_plot(self.scatter)
self.report_caption(caption)
class OWKmeans(OWWidget):
"""
K-means widget
"""
name = "Interactive k-Means"
description = "Widget demonstrates working of k-means algorithm."
icon = "icons/mywidget.svg"
want_main_area = False
# inputs and outputs
inputs = [("Data", Orange.data.Table, "set_data")]
outputs = [("Annotated Data", Table, widget.Default),
("Centroids", Table)]
# settings
numberOfClusters = settings.Setting(1)
autoPlay = False
# data
data = None
# selected attributes in chart
attr_x = settings.Setting('')
attr_y = settings.Setting('')
# other settings
k_means = None
autoPlaySpeed = settings.Setting(1)
lines_to_centroids = settings.Setting(0)
graph_name = 'scatter'
outputName = "cluster"
button_labels = {"step1": "Reassign membership",
"step2": "Recompute centroids",
"step_back": "Step back",
"autoplay_run": "Run",
"autoplay_stop": "Stop",
"random_centroids": "Randomize"}
colors = ['#2f7ed8', '#0d233a', '#8bbc21', '#910000', '#1aadce',
'#492970', '#f28f43', '#77a1e5', '#c42525', '#a6c96a']
def __init__(self):
super().__init__()
# options box
self.optionsBox = gui.widgetBox(self.controlArea)
self.cbx = gui.comboBox(self.optionsBox, self, 'attr_x',
label='X:',
orientation=Qt.Horizontal,
callback=self.restart,
sendSelectedValue=True)
self.cbx.setSizePolicy(QSizePolicy(QSizePolicy.MinimumExpanding, QSizePolicy.Fixed))
self.cby = gui.comboBox(self.optionsBox, self, 'attr_y',
label='Y:',
orientation='horizontal',
callback=self.restart,
sendSelectedValue=True)
self.cby.setSizePolicy(QSizePolicy(QSizePolicy.MinimumExpanding, QSizePolicy.Fixed))
self.centroidsBox = gui.widgetBox(self.controlArea, "Centroids")
self.centroidNumbersSpinner = gui.spin(self.centroidsBox,
self,
'numberOfClusters',
minv=1,
maxv=10,
step=1,
label='Number of centroids:',
callback=self.number_of_clusters_change)
self.centroidNumbersSpinner.setSizePolicy(QSizePolicy(QSizePolicy.MinimumExpanding, QSizePolicy.Fixed))
self.restartButton = gui.button(self.centroidsBox, self, self.button_labels["random_centroids"],
callback=self.restart)
self.linesCheckbox = gui.checkBox(self.centroidsBox,
self,
'lines_to_centroids',
'Show membership lines',
callback=self.complete_replot)
# control box
self.commandsBox = gui.widgetBox(self.controlArea)
self.stepButton = gui.button(self.commandsBox, self, self.button_labels["step2"],
callback=self.step)
self.stepBackButton = gui.button(self.commandsBox, self, self.button_labels["step_back"],
callback=self.step_back)
self.autoPlayButton = gui.button(self.commandsBox, self, self.button_labels["autoplay_run"],
callback=self.auto_play)
self.autoPlaySpeedSpinner = gui.hSlider(self.commandsBox,
self,
'autoPlaySpeed',
minValue=0,
maxValue=1.91,
step=0.1,
intOnly=False,
createLabel=False,
label='Speed:')
gui.rubber(self.controlArea)
# disable until data loaded
self.set_disabled_all(True)
# graph in mainArea
self.scatter = Scatterplot(click_callback=self.graph_clicked,
drop_callback=self.centroid_dropped,
xAxis_gridLineWidth=0,
yAxis_gridLineWidth=0,
title_text='',
tooltip_shared=False,
debug=True) # TODO: set false when end of development
# Just render an empty chart so it shows a nice 'No data to display'
self.scatter.chart()
self.mainArea.layout().addWidget(self.scatter)
def concat_x_y(self):
"""
Function takes two selected columns from data table and merge them in new Orange.data.Table
:return: table with selected columns
:type: Orange.data.Table
"""
attr_x, attr_y = self.data.domain[self.attr_x], self.data.domain[self.attr_y]
cols = []
for attr in (attr_x, attr_y):
subset = self.data[:, attr]
cols.append(subset.Y if subset.Y.size else subset.X)
x = np.column_stack(cols)
domain = Domain([attr_x, attr_y])
return Table(domain, x)
def set_empty_plot(self):
self.scatter.clear()
def set_disabled_all(self, disabled):
"""
Function disable all controls
"""
self.optionsBox.setDisabled(disabled)
self.centroidsBox.setDisabled(disabled)
self.commandsBox.setDisabled(disabled)
def set_data(self, data):
"""
Function receives data from input and init part of widget if data are ok. Otherwise set empty plot and notice
user about that
:param data: input data
:type data: Orange.data.Table or None
"""
self.data = data
def reset_combos():
self.cbx.clear()
self.cby.clear()
def init_combos():
"""
function initialize the combos with attributes
"""
reset_combos()
for var in data.domain if data is not None else []:
if var.is_primitive() and var.is_continuous:
self.cbx.addItem(gui.attributeIconDict[var], var.name)
self.cby.addItem(gui.attributeIconDict[var], var.name)
self.warning(1) # remove warning about too less continuous attributes if exists
self.warning(2) # remove warning about not enough data
if data is None or len(data) == 0:
reset_combos()
self.set_empty_plot()
self.set_disabled_all(True)
elif sum(True for var in data.domain.attributes if isinstance(var, ContinuousVariable)) < 2:
reset_combos()
self.warning(1, "Too few Continuous feature. Min 2 required")
self.set_empty_plot()
self.set_disabled_all(True)
else:
init_combos()
self.set_disabled_all(False)
self.attr_x = self.cbx.itemText(0)
self.attr_y = self.cbx.itemText(1)
if self.k_means is None:
self.k_means = Kmeans(self.concat_x_y())
else:
self.k_means.set_data(self.concat_x_y())
self.modify_kmeans()
def restart(self):
"""
Function triggered on data change or restart button pressed
"""
self.k_means = Kmeans(self.concat_x_y())
self.modify_kmeans()
def modify_kmeans(self):
self.number_of_clusters_change()
self.button_text_change()
def step(self):
"""
Function called on every step
"""
self.k_means.step()
self.replot()
self.button_text_change()
self.send_data()
def step_back(self):
"""
Function called for step back
"""
self.k_means.step_back()
self.replot()
self.button_text_change()
self.send_data()
def button_text_change(self):
"""
Function changes text on ste button and chanbe the button text
"""
self.stepButton.setText(self.button_labels["step2"]
if self.k_means.step_completed
else self.button_labels["step1"])
if self.k_means.stepNo <= 0:
self.stepBackButton.setDisabled(True)
elif not self.autoPlay:
self.stepBackButton.setDisabled(False)
def auto_play(self):
"""
Function called when autoplay button pressed
"""
self.autoPlay = not self.autoPlay
self.autoPlayButton.setText(self.button_labels["autoplay_stop"]
if self.autoPlay
else self.button_labels["autoplay_run"])
if self.autoPlay:
self.optionsBox.setDisabled(True)
self.centroidsBox.setDisabled(True)
self.stepButton.setDisabled(True)
self.stepBackButton.setDisabled(True)
self.autoPlayThread = Autoplay(self)
self.connect(self.autoPlayThread, SIGNAL("step()"), self.step)
self.connect(self.autoPlayThread, SIGNAL("stop_auto_play()"), self.stop_auto_play)
self.autoPlayThread.start()
else:
self.stop_auto_play()
def stop_auto_play(self):
"""
Called when stop autoplay button pressed or in the end of autoplay
"""
self.optionsBox.setDisabled(False)
self.stepButton.setDisabled(False)
self.centroidsBox.setDisabled(False)
self.stepBackButton.setDisabled(False)
self.autoPlay = False
self.autoPlayButton.setText(self.button_labelsp["autoplay_stop"]
if self.autoPlay
else self.button_labels["autoplay_run"])
def replot(self):
"""
Function refreshes the chart
"""
if self.data is None or not self.attr_x or not self.attr_y:
return
if self.k_means.centroids_moved:
# when centroids moved during step
self.scatter.update_series(0, self.k_means.centroids)
if self.lines_to_centroids:
for i, c in enumerate(self.k_means.centroids):
self.scatter.update_series(1 + i, list(chain.from_iterable(
([p[0], p[1]], [c[0], c[1]])
for p in self.k_means.centroids_belonging_points[i])))
else:
self.complete_replot()
def complete_replot(self):
"""
This function performs complete replot of the graph without animation
"""
attr_x, attr_y = self.data.domain[self.attr_x], self.data.domain[self.attr_y]
# plot centroids
options = dict(series=[])
options['series'].append(dict(data=[{'x': p[0],
'y': p[1],
'marker':{'fillColor': self.colors[i % len(self.colors)]}}
for i, p in enumerate(self.k_means.centroids)],
type="scatter",
draggableX=True if self.k_means.step_completed else False,
draggableY=True if self.k_means.step_completed else False,
showInLegend=False,
zIndex=10,
marker=dict(symbol='diamond',
radius=10)))
# plot lines between centroids and points
if self.lines_to_centroids:
for i, c in enumerate(self.k_means.centroids):
options['series'].append(dict(data=list(
chain.from_iterable(([p[0], p[1]], [c[0], c[1]])
for p in self.k_means.centroids_belonging_points[i])),
type="line",
showInLegend=False,
lineWidth=0.2,
enableMouseTracking=False,
color="#ccc"))
# plot data points
for i, points in enumerate(self.k_means.centroids_belonging_points):
options['series'].append(dict(data=points,
type="scatter",
showInLegend=False,
color=rgb_hash_brighter(self.colors[i % len(self.colors)], 30)))
# highcharts parameters
kwargs = dict(
xAxis_title_text=attr_x.name,
yAxis_title_text=attr_y.name,
tooltip_headerFormat="",
tooltip_pointFormat="<strong>%s:</strong> {point.x:.2f} <br/>"
"<strong>%s:</strong> {point.y:.2f}" %
(self.attr_x, self.attr_y))
# plot
self.scatter.chart(options, **kwargs)
def replot_series(self):
"""
This function replot just series connected with centroids and uses animation for that
"""
k = self.k_means.k
series = []
# plot lines between centroids and points
if self.lines_to_centroids:
for i, c in enumerate(self.k_means.centroids):
series.append(dict(
data=list(chain.from_iterable(([p[0], p[1]], [c[0], c[1]])
for p in self.k_means.centroids_belonging_points[i])),
type="line",
showInLegend=False,
lineWidth=0.2,
enableMouseTracking=False,
color="#ccc"))
# plot data points
for i, points in enumerate(self.k_means.centroids_belonging_points):
series.append(dict(data=points,
type="scatter",
showInLegend=False,
color=rgb_hash_brighter(self.colors[i % len(self.colors)], 30)))
self.scatter.add_series(series)
self.scatter.remove_last_series(k * 2 if self.lines_to_centroids else k)
def number_of_clusters_change(self):
"""
Function that change number of clusters if required
"""
if self.numberOfClusters > len(self.data):
# if too less data for clusters number
self.warning(2, "Please provide at least number of points equal to "
"number of clusters selected or decrease number of clusters")
self.set_empty_plot()
self.commandsBox.setDisabled(True)
else:
self.warning(2)
self.commandsBox.setDisabled(False)
if self.k_means is None: # if before too less data k_means is None
self.k_means = Kmeans(self.concat_x_y())
if self.k_means.k < self.numberOfClusters:
self.k_means.add_centroids(self.numberOfClusters - self.k_means.k)
elif not self.k_means.k == self.numberOfClusters:
self.k_means.delete_centroids(self.k_means.k - self.numberOfClusters)
self.replot()
self.send_data()
def graph_clicked(self, x, y):
"""
Function called when user click in graph. Centroid have to be added.
"""
if self.k_means is not None:
self.k_means.add_centroids([x, y])
self.numberOfClusters += 1
self.replot()
self.send_data()
self.button_text_change()
def centroid_dropped(self, _index, x, y):
"""
Function called when centroid with _index moved.
"""
self.k_means.move_centroid(_index, x, y)
self.complete_replot()
self.send_data()
self.button_text_change()
def send_data(self):
"""
Function sends data with clusters column and data with centroids position to the output
"""
if self.k_means is None or self.k_means.clusters is None:
self.send("Annotated Data", None)
self.send("Centroids", None)
else:
clust_var = DiscreteVariable(
self.outputName, values=["C%d" % (x + 1) for x in range(self.k_means.k)])
attributes, classes = self.data.domain.attributes, self.data.domain.class_vars
meta_attrs = self.data.domain.metas
if classes:
meta_attrs += classes
classes = [clust_var]
domain = Domain(attributes, classes, meta_attrs)
annotated_data = Table.from_table(domain, self.data)
annotated_data.get_column_view(clust_var)[0][:] = self.k_means.clusters
centroids = Table(Domain(self.k_means.data.domain.attributes), self.k_means.centroids)
self.send("Annotated Data", annotated_data)
self.send("Centroids", centroids)