Skip to content

SGD: Make normalization optional, fix settings #171

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
janezd merged 2 commits into from
Dec 5, 2025
Merged

SGD: Make normalization optional, fix settings #171

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4f660e2
SGD: Make normalization optional
janezd Dec 5, 2025
6cb5b14
SGD: Fix settings; use hints
janezd Dec 5, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
67 changes: 45 additions & 22 deletions orangecontrib/educational/widgets/owgradientdescent.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@
from Orange.preprocess.transformation import Indicator
from Orange.widgets import gui
from Orange.widgets import settings
from Orange.widgets.utils.itemmodels import DomainModel
from Orange.widgets.utils.itemmodels import VariableListModel
from Orange.widgets.widget import OWWidget, Msg, Input, Output
from Orange.preprocess.preprocess import Normalize
from orangewidget.report import report
Expand Down Expand Up @@ -94,11 +94,10 @@ class Outputs:

graph_name = "graph" # QGraphicsView (pg.PlotWidget)

settingsHandler = settings.DomainContextHandler(
match_values=settings.DomainContextHandler.MATCH_VALUES_CLASS)
attr_x = settings.ContextSetting(None)
attr_y = settings.ContextSetting(None)
target_class = settings.Setting('')
attr_x_hint = settings.Setting("")
attr_y_hint = settings.Setting("")
normalization = settings.Setting(True)
target_class_hint = settings.Setting('')
alpha = settings.Setting(0.4)
step_size = settings.Setting(30) # step size for stochastic gds
auto_play_speed = settings.Setting(1)
Expand All @@ -121,6 +120,8 @@ def __init__(self):
super().__init__()
self.learner = None
self.data = None
self.attr_x = self.attr_y = None
self.target_class = ""
self.selected_data = None
self.cost_grid = None
self.min_x = self.max_x = self.min_y = self.max_y = None
Expand All @@ -132,15 +133,17 @@ def __init__(self):
self.auto_play_button_text = ["Run", "Stop"]
self.auto_play_thread = None

self.var_model = DomainModel(valid_types=ContinuousVariable,
order=DomainModel.ATTRIBUTES)
self.var_model = VariableListModel()

self.options_box = gui.widgetBox(self.controlArea, "Variables")
opts = dict(
widget=self.options_box, master=self, orientation=Qt.Horizontal,
callback=self.change_attributes)
gui.comboBox(value='attr_x', model=self.var_model, **opts)
gui.comboBox(value='attr_y', model=self.var_model, **opts)
gui.checkBox(self.options_box, self, 'normalization',
'Scale and center data',
callback=self.change_attributes)
gui.comboBox(value='target_class', label='Target: ',
sendSelectedValue=True, **opts)

Expand Down Expand Up @@ -207,6 +210,9 @@ def _add_graph(self):
# User-interaction

def change_attributes(self):
self.attr_x_hint = self.attr_x.name if self.attr_x else ""
self.attr_y_hint = self.attr_y.name if self.attr_y else ""
self.target_class_hint = self.target_class
self.set_axis_titles()
self.learner = None # that theta does not same equal
self.restart()
Expand Down Expand Up @@ -323,25 +329,26 @@ def keyPressEvent(self, e):
@Inputs.data
def set_data(self, data):
target_combo = self.controls.target_class

self.closeContext()
self.Error.clear()

self.cost_grid = None
self.learner = None
self.selected_data = None
self.data = None
domain = None # for pylint
target_combo.clear()
self.clear_plot()

if data:
self.var_model.set_domain(data.domain)
domain = data.domain
self.var_model[:] = [
attr for attr in domain.attributes if attr.is_continuous]
if data.domain.class_var is None:
data = None
self.Error.no_class()
# don't use self.is_classification -- widget has no self.data yet
elif data.domain.class_var.is_discrete:
if len(data.domain.class_var.values) < 2:
elif domain.class_var.is_discrete:
if len(domain.class_var.values) < 2:
self.Error.no_class_values()
data = None
elif len(self.var_model) < 2:
Expand All @@ -352,24 +359,37 @@ def set_data(self, data):
data = None

if not data:
self.var_model.set_domain(None)
self.var_model.clear()
self.send_output()
return

assert domain is not None
self.data = data
self.controls.attr_y.setHidden(not self.is_classification)
target_combo.box.setHidden(not self.is_classification)
self.attr_x = self.var_model[0]
attr_names = [var.name for var in self.var_model]
if self.attr_x_hint in attr_names:
self.attr_x = domain[self.attr_x_hint]
else:
self.attr_x = self.var_model[0]
self.attr_x_hint = self.attr_x.name
if self.is_classification:
self.attr_y = self.var_model[min(1, len(self.var_model))]
if self.attr_y_hint in attr_names:
self.attr_y = domain[self.attr_y_hint]
else:
self.attr_y = self.var_model[int(self.attr_x == self.var_model[0])]
self.attr_y_hint = self.attr_y.name
target_combo.clear()
values = data.domain.class_var.values
target_combo.addItems(values)
# For binary class, take value with index 1 as a target, so the
# output class is the same as the input
self.target_class = values[1 if len(values) == 2 else 0]
# output class is the same as the inputđ
if self.target_class_hint in values:
self.target_class = self.target_class_hint
else:
self.target_class = values[1 if len(values) == 2 else 0]
self.target_class_hint = self.target_class
self.step_size_spin.setMaximum(len(data))
self.openContext(self.data)
self.set_axis_titles()
self.restart()

Expand Down Expand Up @@ -413,9 +433,12 @@ def select_columns(self):
return

data = data[valid_data]
self.selected_data = Normalize(
transform_class=not self.is_classification)(
data)
if self.normalization:
self.selected_data = Normalize(
transform_class=not self.is_classification)(
data)
else:
self.selected_data = data

def send_output(self):
self.send_model()
Expand Down
67 changes: 67 additions & 0 deletions orangecontrib/educational/widgets/tests/test_owgradientdescent.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,7 @@

from orangecontrib.educational.widgets.owgradientdescent import \
OWGradientDescent, GRID_SIZE
from orangewidget.tests.utils import simulate


class TestOWGradientDescent(WidgetTest):
Expand Down Expand Up @@ -147,6 +148,59 @@ def test_set_data_continuous(self):
self.assertFalse(w.controls.target_class.isVisible())
self.assertIs(w.attr_x, continuous_attributes[0])

def test_attr_hints(self):
w = self.widget
combo_x, combo_y = w.controls.attr_x, w.controls.attr_y
combo_t = w.controls.target_class
a, b = ContinuousVariable("a"), ContinuousVariable("b")
c = DiscreteVariable("c", values=("a", "b", "c"))
data = Table.from_numpy(Domain([a, b], c), [[1, 2], [2, 3]], [0, 1])

self.send_signal(w.Inputs.data, data)
self.assertEqual(len(w.var_model), 2)
self.assertFalse(w.controls.target_class.box.isHidden())
self.assertFalse(w.controls.attr_y.isHidden())

simulate.combobox_activate_index(combo_x, 1)
simulate.combobox_activate_index(combo_y, 0)
simulate.combobox_activate_item(combo_t, c.values[1])

self.send_signal(w.Inputs.data, None)
self.assertEqual(len(w.var_model), 0)

self.send_signal(w.Inputs.data, data)
self.assertEqual(len(w.var_model), 2)
self.assertIs(w.attr_x, b)
self.assertIs(w.attr_y, a)
self.assertEqual(w.target_class, c.values[1])

self.send_signal(w.Inputs.data, data.transform(Domain([a, c], b)))
self.assertEqual(len(w.var_model), 1)
self.assertTrue(w.controls.attr_y.isHidden())
self.assertTrue(w.controls.target_class.box.isHidden())

self.send_signal(w.Inputs.data, None)
self.assertEqual(len(w.var_model), 0)

self.send_signal(w.Inputs.data, data.transform(Domain([a], b)))
self.assertEqual(len(w.var_model), 1)

d, e = ContinuousVariable("d"), ContinuousVariable("e")
data = Table.from_numpy(
Domain([d, a, e], b),
[[1, 2, 5], [2, 3, 9]], [0.1, 1])
self.send_signal(w.Inputs.data, data)
self.assertEqual(len(w.var_model), 3)
self.assertIs(w.attr_x, a)

simulate.combobox_activate_index(combo_x, 2)
self.assertIs(w.attr_x, e)

self.send_signal(w.Inputs.data, None)
self.assertEqual(len(w.var_model), 0)
self.send_signal(w.Inputs.data, data)
self.assertIs(w.attr_x, e)

def test_set_learner_discrete(self):
"""
Test if restart works fine
Expand Down Expand Up @@ -189,6 +243,19 @@ def test_learner_set_continuous(self):
self.assertEqual(w.learner.stochastic, False)
self.assertEqual(w.learner.stochastic_step_size, w.step_size)

def test_normalization(self):
w = self.widget
w.normalization = False
a, b = ContinuousVariable("a"), ContinuousVariable("b")
c = DiscreteVariable("c", values=("a", "b", "c"))
data = Table.from_numpy(Domain([a, b], c), [[1, 2], [2, 3]], [0, 1])
self.send_signal(w.Inputs.data, data)

np.testing.assert_equal(w.selected_data.X, data.X)

w.controls.normalization.click()
np.testing.assert_equal(w.selected_data.X, [[-1, -1], [1, 1]])

def test_learner_set_continuous_restart(self):
"""
Test if restart works fine
Expand Down
Loading