def paint(self, painter, option, widget=None):
brush = self.brush()
if self.__hover:
brush = QBrush(brush.color().darker(110))
painter.setBrush(brush)
painter.setPen(self.pen())
painter.drawEllipse(self.rect())
def brush_darker(brush, factor):
"""Return a copy of the brush darkened by factor.
"""
grad = brush.gradient()
if grad:
return QBrush(gradient_darker(grad, factor))
else:
brush = QBrush(brush)
brush.setColor(brush.color().darker(factor))
return brush
def paint(self, painter, option, widget=None):
painter.save()
path = self.path()
brush = QBrush(self.brush())
pen = QPen(self.pen())
if option.state & QStyle.State_Selected:
pen.setColor(Qt.red)
brush.setStyle(Qt.DiagCrossPattern)
brush.setColor(QColor(40, 40, 40, 100))
elif option.state & QStyle.State_MouseOver:
pen.setColor(Qt.blue)
if option.state & QStyle.State_MouseOver:
brush.setColor(QColor(100, 100, 100, 100))
if brush.style() == Qt.NoBrush:
# Make sure the highlight is actually visible.
brush.setStyle(Qt.SolidPattern)
painter.setPen(pen)
painter.setBrush(brush)
painter.drawPath(path)
painter.restore()
def color_for_label(self, ind, light=100):
if self.label_colors is None:
return Qt.lightGray
return QBrush(self.label_colors[ind].lighter(light))
def setBrush(self, brush):
brush = QBrush(brush)
if self.__brush != brush:
self.__brush = brush
for item in self.__items:
item.setBrush(brush)
def _update(self):
def _isinvalid(x):
return isnan(x) or isinf(x)
# Update the displayed confusion matrix
if self.results is not None and self.selected_learner:
cmatrix = confusion_matrix(self.results, self.selected_learner[0])
colsum = cmatrix.sum(axis=0)
rowsum = cmatrix.sum(axis=1)
n = len(cmatrix)
diag = numpy.diag_indices(n)
colors = cmatrix.astype(numpy.double)
colors[diag] = 0
if self.selected_quantity == 0:
normalized = cmatrix.astype(numpy.int)
formatstr = "{}"
div = numpy.array([colors.max()])
else:
if self.selected_quantity == 1:
normalized = 100 * cmatrix / colsum
div = colors.max(axis=0)
else:
normalized = 100 * cmatrix / rowsum[:, numpy.newaxis]
div = colors.max(axis=1)[:, numpy.newaxis]
formatstr = "{:2.1f} %"
div[div == 0] = 1
colors /= div
colors[diag] = normalized[diag] / normalized[diag].max()
for i in range(n):
for j in range(n):
val = normalized[i, j]
col_val = colors[i, j]
item = self._item(i + 2, j + 2)
item.setData(
"NA" if _isinvalid(val) else formatstr.format(val),
Qt.DisplayRole)
bkcolor = QColor.fromHsl(
[0, 240][i == j], 160,
255 if _isinvalid(col_val) else int(255 -
30 * col_val))
item.setData(QBrush(bkcolor), Qt.BackgroundRole)
item.setData("trbl", BorderRole)
item.setToolTip("actual: {}\npredicted: {}".format(
self.headers[i], self.headers[j]))
item.setTextAlignment(Qt.AlignRight | Qt.AlignVCenter)
item.setFlags(Qt.ItemIsEnabled | Qt.ItemIsSelectable)
self._set_item(i + 2, j + 2, item)
bold_font = self.tablemodel.invisibleRootItem().font()
bold_font.setBold(True)
def _sum_item(value, border=""):
item = QStandardItem()
item.setData(value, Qt.DisplayRole)
item.setTextAlignment(Qt.AlignRight | Qt.AlignVCenter)
item.setFlags(Qt.ItemIsEnabled)
item.setFont(bold_font)
item.setData(border, BorderRole)
item.setData(QColor(192, 192, 192), BorderColorRole)
return item
for i in range(n):
self._set_item(n + 2, i + 2, _sum_item(int(colsum[i]), "t"))
self._set_item(i + 2, n + 2, _sum_item(int(rowsum[i]), "l"))
self._set_item(n + 2, n + 2, _sum_item(int(rowsum.sum())))
def _setup_plot(self):
def merge_averaging():
for curve in curves:
graphics = curve.merge()
curve = graphics.curve
self.plot.addItem(graphics.curve_item)
if self.display_convex_curve:
self.plot.addItem(graphics.hull_item)
if self.display_def_threshold and curve.is_valid:
points = curve.points
ind = numpy.argmin(numpy.abs(points.thresholds - 0.5))
item = pg.TextItem(text="{:.3f}".format(
points.thresholds[ind]), )
item.setPos(points.fpr[ind], points.tpr[ind])
self.plot.addItem(item)
hull_curves = [curve.merged.hull for curve in selected]
if hull_curves:
self._rocch = convex_hull(hull_curves)
iso_pen = QPen(QColor(Qt.black), 1)
iso_pen.setCosmetic(True)
self._perf_line = InfiniteLine(pen=iso_pen, antialias=True)
self.plot.addItem(self._perf_line)
return hull_curves
def vertical_averaging():
for curve in curves:
graphics = curve.avg_vertical()
self.plot.addItem(graphics.curve_item)
self.plot.addItem(graphics.confint_item)
return [curve.avg_vertical.hull for curve in selected]
def threshold_averaging():
for curve in curves:
graphics = curve.avg_threshold()
self.plot.addItem(graphics.curve_item)
self.plot.addItem(graphics.confint_item)
return [curve.avg_threshold.hull for curve in selected]
def no_averaging():
for curve in curves:
graphics = curve.folds()
for fold in graphics:
self.plot.addItem(fold.curve_item)
if self.display_convex_curve:
self.plot.addItem(fold.hull_item)
return [fold.hull for curve in selected for fold in curve.folds]
averagings = {
OWROCAnalysis.Merge: merge_averaging,
OWROCAnalysis.Vertical: vertical_averaging,
OWROCAnalysis.Threshold: threshold_averaging,
OWROCAnalysis.NoAveraging: no_averaging
}
target = self.target_index
selected = self.selected_classifiers
curves = [self.plot_curves(target, i) for i in selected]
selected = [self.curve_data(target, i) for i in selected]
hull_curves = averagings[self.roc_averaging]()
if self.display_convex_hull and hull_curves:
hull = convex_hull(hull_curves)
hull_pen = QPen(QColor(200, 200, 200, 100), 2)
hull_pen.setCosmetic(True)
item = self.plot.plot(hull.fpr,
hull.tpr,
pen=hull_pen,
brush=QBrush(QColor(200, 200, 200, 50)),
fillLevel=0)
item.setZValue(-10000)
pen = QPen(QColor(100, 100, 100, 100), 1, Qt.DashLine)
pen.setCosmetic(True)
self.plot.plot([0, 1], [0, 1], pen=pen, antialias=True)
if self.roc_averaging == OWROCAnalysis.Merge:
self._update_perf_line()
warning = ""
if not all(c.is_valid for c in hull_curves):
if any(c.is_valid for c in hull_curves):
warning = "Some ROC curves are undefined"
else:
warning = "All ROC curves are undefined"
self.warning(warning)
class OWSOM(OWWidget):
name = "Self-Organizing Map"
description = "Computation of self-organizing map."
icon = "icons/SOM.svg"
keywords = ["SOM"]
class Inputs:
data = Input("Data", Table)
class Outputs:
selected_data = Output("Selected Data", Table, default=True)
annotated_data = Output(ANNOTATED_DATA_SIGNAL_NAME, Table)
settingsHandler = DomainContextHandler()
auto_dimension = Setting(True)
size_x = Setting(10)
size_y = Setting(10)
hexagonal = Setting(1)
initialization = Setting(0)
attr_color = ContextSetting(None)
size_by_instances = Setting(True)
pie_charts = Setting(False)
selection = Setting(None, schema_only=True)
graph_name = "view"
_grid_pen = QPen(QBrush(QColor(224, 224, 224)), 2)
_grid_pen.setCosmetic(True)
OptControls = namedtuple(
"OptControls",
("shape", "auto_dim", "spin_x", "spin_y", "initialization", "start"))
class Warning(OWWidget.Warning):
ignoring_disc_variables = Msg("SOM ignores categorical variables.")
missing_colors = \
Msg("Some data instances have undefined value of '{}'.")
missing_values = \
Msg("{} data instance{} with undefined value(s) {} not shown.")
single_attribute = Msg("Data contains a single numeric column.")
class Error(OWWidget.Error):
no_numeric_variables = Msg("Data contains no numeric columns.")
no_defined_rows = Msg("All rows contain at least one undefined value.")
def __init__(self):
super().__init__()
self.__pending_selection = self.selection
self._optimizer = None
self._optimizer_thread = None
self.stop_optimization = False
self.data = self.cont_x = None
self.cells = self.member_data = None
self.selection = None
self.colors = self.thresholds = self.bin_labels = None
box = gui.vBox(self.controlArea, box="SOM")
shape = gui.comboBox(box,
self,
"",
items=("Hexagonal grid", "Square grid"))
shape.setCurrentIndex(1 - self.hexagonal)
box2 = gui.indentedBox(box, 10)
auto_dim = gui.checkBox(box2,
self,
"auto_dimension",
"Set dimensions automatically",
callback=self.on_auto_dimension_changed)
self.manual_box = box3 = gui.hBox(box2)
spinargs = dict(value="",
widget=box3,
master=self,
minv=5,
maxv=100,
step=5,
alignment=Qt.AlignRight)
spin_x = gui.spin(**spinargs)
spin_x.setValue(self.size_x)
gui.widgetLabel(box3, "×")
spin_y = gui.spin(**spinargs)
spin_y.setValue(self.size_y)
gui.rubber(box3)
self.manual_box.setEnabled(not self.auto_dimension)
initialization = gui.comboBox(box,
self,
"initialization",
items=("Initialize with PCA",
"Random initialization",
"Replicable random"))
start = gui.button(box,
self,
"Restart",
callback=self.restart_som_pressed,
sizePolicy=(QSizePolicy.MinimumExpanding,
QSizePolicy.Fixed))
self.opt_controls = self.OptControls(shape, auto_dim, spin_x, spin_y,
initialization, start)
box = gui.vBox(self.controlArea, "Color")
gui.comboBox(box,
self,
"attr_color",
searchable=True,
callback=self.on_attr_color_change,
model=DomainModel(placeholder="(Same color)",
valid_types=DomainModel.PRIMITIVE))
gui.checkBox(box,
self,
"pie_charts",
label="Show pie charts",
callback=self.on_pie_chart_change)
gui.checkBox(box,
self,
"size_by_instances",
label="Size by number of instances",
callback=self.on_attr_size_change)
gui.rubber(self.controlArea)
self.scene = QGraphicsScene(self)
self.view = SomView(self.scene)
self.view.setMinimumWidth(400)
self.view.setMinimumHeight(400)
self.view.setHorizontalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.view.setVerticalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.view.setRenderHint(QPainter.Antialiasing)
self.view.selection_changed.connect(self.on_selection_change)
self.view.selection_moved.connect(self.on_selection_move)
self.view.selection_mark_changed.connect(self.on_selection_mark_change)
self.mainArea.layout().addWidget(self.view)
self.elements = None
self.grid = None
self.grid_cells = None
self.legend = None
@Inputs.data
def set_data(self, data):
def prepare_data():
if len(cont_attrs) < len(attrs):
self.Warning.ignoring_disc_variables()
if len(cont_attrs) == 1:
self.Warning.single_attribute()
x = Table.from_table(Domain(cont_attrs), data).X
if sp.issparse(x):
self.data = data
self.cont_x = x.tocsr()
else:
mask = np.all(np.isfinite(x), axis=1)
if not np.any(mask):
self.Error.no_defined_rows()
else:
if np.all(mask):
self.data = data
self.cont_x = x.copy()
else:
self.data = data[mask]
self.cont_x = x[mask]
self.cont_x -= np.min(self.cont_x, axis=0)[None, :]
sums = np.sum(self.cont_x, axis=0)[None, :]
sums[sums == 0] = 1
self.cont_x /= sums
def set_warnings():
missing = len(data) - len(self.data)
if missing == 1:
self.Warning.missing_values(1, "", "is")
elif missing > 1:
self.Warning.missing_values(missing, "s", "are")
self.stop_optimization_and_wait()
self.closeContext()
self.clear()
self.Error.clear()
self.Warning.clear()
if data is not None:
attrs = data.domain.attributes
cont_attrs = [var for var in attrs if var.is_continuous]
if not cont_attrs:
self.Error.no_numeric_variables()
else:
prepare_data()
if self.data is not None:
self.controls.attr_color.model().set_domain(data.domain)
self.attr_color = data.domain.class_var
set_warnings()
self.openContext(self.data)
self.set_color_bins()
self.create_legend()
self.recompute_dimensions()
self.start_som()
def clear(self):
self.data = self.cont_x = None
self.cells = self.member_data = None
self.attr_color = None
self.colors = self.thresholds = self.bin_labels = None
if self.elements is not None:
self.scene.removeItem(self.elements)
self.elements = None
self.clear_selection()
self.controls.attr_color.model().set_domain(None)
self.Warning.clear()
self.Error.clear()
def recompute_dimensions(self):
if not self.auto_dimension or self.cont_x is None:
return
dim = max(5, int(np.ceil(np.sqrt(5 * np.sqrt(self.cont_x.shape[0])))))
self.opt_controls.spin_x.setValue(dim)
self.opt_controls.spin_y.setValue(dim)
def on_auto_dimension_changed(self):
self.manual_box.setEnabled(not self.auto_dimension)
if self.auto_dimension:
self.recompute_dimensions()
else:
spin_x = self.opt_controls.spin_x
spin_y = self.opt_controls.spin_y
dimx = int(5 * np.round(spin_x.value() / 5))
dimy = int(5 * np.round(spin_y.value() / 5))
spin_x.setValue(dimx)
spin_y.setValue(dimy)
def on_attr_color_change(self):
self.controls.pie_charts.setEnabled(self.attr_color is not None)
self.set_color_bins()
self.create_legend()
self.rescale()
self._redraw()
def on_attr_size_change(self):
self._redraw()
def on_pie_chart_change(self):
self._redraw()
def clear_selection(self):
self.selection = None
self.redraw_selection()
def on_selection_change(self, selection, action=SomView.SelectionSet):
if self.data is None: # clicks on empty canvas
return
if self.selection is None:
self.selection = np.zeros(self.grid_cells.T.shape, dtype=np.int16)
if action == SomView.SelectionSet:
self.selection[:] = 0
self.selection[selection] = 1
elif action == SomView.SelectionAddToGroup:
self.selection[selection] = max(1, np.max(self.selection))
elif action == SomView.SelectionNewGroup:
self.selection[selection] = 1 + np.max(self.selection)
elif action & SomView.SelectionRemove:
self.selection[selection] = 0
self.redraw_selection()
self.update_output()
def on_selection_move(self, event: QKeyEvent):
if self.selection is None or not np.any(self.selection):
if event.key() in (Qt.Key_Right, Qt.Key_Down):
x = y = 0
else:
x = self.size_x - 1
y = self.size_y - 1
else:
x, y = np.nonzero(self.selection)
if len(x) > 1:
return
if event.key() == Qt.Key_Up and y > 0:
y -= 1
if event.key() == Qt.Key_Down and y < self.size_y - 1:
y += 1
if event.key() == Qt.Key_Left and x:
x -= 1
if event.key() == Qt.Key_Right and x < self.size_x - 1:
x += 1
x -= self.hexagonal and x == self.size_x - 1 and y % 2
if self.selection is not None and self.selection[x, y]:
return
selection = np.zeros(self.grid_cells.shape, dtype=bool)
selection[x, y] = True
self.on_selection_change(selection)
def on_selection_mark_change(self, marks):
self.redraw_selection(marks=marks)
def redraw_selection(self, marks=None):
if self.grid_cells is None:
return
sel_pen = QPen(QBrush(QColor(128, 128, 128)), 2)
sel_pen.setCosmetic(True)
mark_pen = QPen(QBrush(QColor(128, 128, 128)), 4)
mark_pen.setCosmetic(True)
pens = [self._grid_pen, sel_pen]
mark_brush = QBrush(QColor(224, 255, 255))
sels = self.selection is not None and np.max(self.selection)
palette = LimitedDiscretePalette(number_of_colors=sels + 1)
brushes = [QBrush(Qt.NoBrush)] + \
[QBrush(palette[i].lighter(165)) for i in range(sels)]
for y in range(self.size_y):
for x in range(self.size_x - (y % 2) * self.hexagonal):
cell = self.grid_cells[y, x]
marked = marks is not None and marks[x, y]
sel_group = self.selection is not None and self.selection[x, y]
if marked:
cell.setBrush(mark_brush)
cell.setPen(mark_pen)
else:
cell.setBrush(brushes[sel_group])
cell.setPen(pens[bool(sel_group)])
cell.setZValue(marked or sel_group)
def restart_som_pressed(self):
if self._optimizer_thread is not None:
self.stop_optimization = True
self._optimizer.stop_optimization = True
else:
self.start_som()
def start_som(self):
self.read_controls()
self.update_layout()
self.clear_selection()
if self.cont_x is not None:
self.enable_controls(False)
self._recompute_som()
else:
self.update_output()
def read_controls(self):
c = self.opt_controls
self.hexagonal = c.shape.currentIndex() == 0
self.size_x = c.spin_x.value()
self.size_y = c.spin_y.value()
def enable_controls(self, enable):
c = self.opt_controls
c.shape.setEnabled(enable)
c.auto_dim.setEnabled(enable)
c.start.setText("Start" if enable else "Stop")
def update_layout(self):
self.set_legend_pos()
if self.elements: # Prevent having redrawn grid but with old elements
self.scene.removeItem(self.elements)
self.elements = None
self.redraw_grid()
self.rescale()
def _redraw(self):
self.Warning.missing_colors.clear()
if self.elements:
self.scene.removeItem(self.elements)
self.elements = None
self.view.set_dimensions(self.size_x, self.size_y, self.hexagonal)
if self.cells is None:
return
sizes = self.cells[:, :, 1] - self.cells[:, :, 0]
sizes = sizes.astype(float)
if not self.size_by_instances:
sizes[sizes != 0] = 0.8
else:
sizes *= 0.8 / np.max(sizes)
self.elements = QGraphicsItemGroup()
self.scene.addItem(self.elements)
if self.attr_color is None:
self._draw_same_color(sizes)
elif self.pie_charts:
self._draw_pie_charts(sizes)
else:
self._draw_colored_circles(sizes)
@property
def _grid_factors(self):
return (0.5, sqrt3_2) if self.hexagonal else (0, 1)
def _draw_same_color(self, sizes):
fx, fy = self._grid_factors
color = QColor(64, 64, 64)
for y in range(self.size_y):
for x in range(self.size_x - self.hexagonal * (y % 2)):
r = sizes[x, y]
n = len(self.get_member_indices(x, y))
if not r:
continue
ellipse = ColoredCircle(r / 2, color, 0)
ellipse.setPos(x + (y % 2) * fx, y * fy)
ellipse.setToolTip(f"{n} instances")
self.elements.addToGroup(ellipse)
def _get_color_column(self):
color_column = \
self.data.get_column_view(self.attr_color)[0].astype(float,
copy=False)
if self.attr_color.is_discrete:
with np.errstate(invalid="ignore"):
int_col = color_column.astype(int)
int_col[np.isnan(color_column)] = len(self.colors)
else:
int_col = np.zeros(len(color_column), dtype=int)
# The following line is unnecessary because rows with missing
# numeric data are excluded. Uncomment it if you change SOM to
# tolerate missing values.
# int_col[np.isnan(color_column)] = len(self.colors)
for i, thresh in enumerate(self.thresholds, start=1):
int_col[color_column >= thresh] = i
return int_col
def _tooltip(self, colors, distribution):
if self.attr_color.is_discrete:
values = self.attr_color.values
else:
values = self._bin_names()
tot = np.sum(distribution)
nbhp = "\N{NON-BREAKING HYPHEN}"
return '<table style="white-space: nowrap">' + "".join(f"""
<tr>
<td>
<font color={color.name()}>■</font>
<b>{escape(val).replace("-", nbhp)}</b>:
</td>
<td>
{n} ({n / tot * 100:.1f} %)
</td>
</tr>
""" for color, val, n in zip(colors, values, distribution) if n) \
+ "</table>"
def _draw_pie_charts(self, sizes):
fx, fy = self._grid_factors
color_column = self._get_color_column()
colors = self.colors.qcolors_w_nan
for y in range(self.size_y):
for x in range(self.size_x - self.hexagonal * (y % 2)):
r = sizes[x, y]
if not r:
self.grid_cells[y, x].setToolTip("")
continue
members = self.get_member_indices(x, y)
color_dist = np.bincount(color_column[members],
minlength=len(colors))
rel_color_dist = color_dist.astype(float) / len(members)
pie = PieChart(rel_color_dist, r / 2, colors)
pie.setToolTip(self._tooltip(colors, color_dist))
self.elements.addToGroup(pie)
pie.setPos(x + (y % 2) * fx, y * fy)
def _draw_colored_circles(self, sizes):
fx, fy = self._grid_factors
color_column = self._get_color_column()
qcolors = self.colors.qcolors_w_nan
for y in range(self.size_y):
for x in range(self.size_x - self.hexagonal * (y % 2)):
r = sizes[x, y]
if not r:
continue
members = self.get_member_indices(x, y)
color_dist = color_column[members]
color_dist = color_dist[color_dist < len(self.colors)]
if len(color_dist) != len(members):
self.Warning.missing_colors(self.attr_color.name)
bc = np.bincount(color_dist, minlength=len(self.colors))
color = qcolors[np.argmax(bc)]
ellipse = ColoredCircle(r / 2, color,
np.max(bc) / len(members))
ellipse.setPos(x + (y % 2) * fx, y * fy)
ellipse.setToolTip(self._tooltip(qcolors, bc))
self.elements.addToGroup(ellipse)
def redraw_grid(self):
if self.grid is not None:
self.scene.removeItem(self.grid)
self.grid = QGraphicsItemGroup()
self.grid.setZValue(-200)
self.grid_cells = np.full((self.size_y, self.size_x), None)
for y in range(self.size_y):
for x in range(self.size_x - (y % 2) * self.hexagonal):
if self.hexagonal:
cell = QGraphicsPathItem(_hexagon_path)
cell.setPos(x + (y % 2) / 2, y * sqrt3_2)
else:
cell = QGraphicsRectItem(x - 0.5, y - 0.5, 1, 1)
self.grid_cells[y, x] = cell
cell.setPen(self._grid_pen)
self.grid.addToGroup(cell)
self.scene.addItem(self.grid)
def get_member_indices(self, x, y):
i, j = self.cells[x, y]
return self.member_data[i:j]
def _recompute_som(self):
if self.cont_x is None:
return
som = SOM(self.size_x,
self.size_y,
hexagonal=self.hexagonal,
pca_init=self.initialization == 0,
random_seed=0 if self.initialization == 2 else None)
class Optimizer(QObject):
update = Signal(float, np.ndarray, np.ndarray)
done = Signal(SOM)
stopped = Signal()
stop_optimization = False
def __init__(self, data, som):
super().__init__()
self.som = som
self.data = data
def callback(self, progress):
self.update.emit(progress, self.som.weights.copy(),
self.som.ssum_weights.copy())
return not self.stop_optimization
def run(self):
try:
self.som.fit(self.data,
N_ITERATIONS,
callback=self.callback)
# Report an exception, but still remove the thread
finally:
self.done.emit(self.som)
self.stopped.emit()
def thread_finished():
self._optimizer = None
self._optimizer_thread = None
self.progressBarInit()
self._optimizer = Optimizer(self.cont_x, som)
self._optimizer_thread = QThread()
self._optimizer_thread.setStackSize(5 * 2**20)
self._optimizer.update.connect(self.__update)
self._optimizer.done.connect(self.__done)
self._optimizer.stopped.connect(self._optimizer_thread.quit)
self._optimizer.moveToThread(self._optimizer_thread)
self._optimizer_thread.started.connect(self._optimizer.run)
self._optimizer_thread.finished.connect(thread_finished)
self.stop_optimization = False
self._optimizer_thread.start()
@Slot(float, object, object)
def __update(self, _progress, weights, ssum_weights):
self.progressBarSet(_progress)
self._assign_instances(weights, ssum_weights)
self._redraw()
@Slot(object)
def __done(self, som):
self.enable_controls(True)
self.progressBarFinished()
self._assign_instances(som.weights, som.ssum_weights)
self._redraw()
# This is the first time we know what was selected (assuming that
# initialization is not set to random)
if self.__pending_selection is not None:
self.on_selection_change(self.__pending_selection)
self.__pending_selection = None
self.update_output()
def stop_optimization_and_wait(self):
if self._optimizer_thread is not None:
self.stop_optimization = True
self._optimizer.stop_optimization = True
self._optimizer_thread.quit()
self._optimizer_thread.wait()
self._optimizer_thread = None
def onDeleteWidget(self):
self.stop_optimization_and_wait()
self.clear()
super().onDeleteWidget()
def _assign_instances(self, weights, ssum_weights):
if self.cont_x is None:
return # the widget is shutting down while signals still processed
assignments = SOM.winner_from_weights(self.cont_x, weights,
ssum_weights, self.hexagonal)
members = defaultdict(list)
for i, (x, y) in enumerate(assignments):
members[(x, y)].append(i)
members.pop(None, None)
self.cells = np.empty((self.size_x, self.size_y, 2), dtype=int)
self.member_data = np.empty(self.cont_x.shape[0], dtype=int)
index = 0
for x in range(self.size_x):
for y in range(self.size_y):
nmembers = len(members[(x, y)])
self.member_data[index:index + nmembers] = members[(x, y)]
self.cells[x, y] = [index, index + nmembers]
index += nmembers
def resizeEvent(self, event):
super().resizeEvent(event)
self.create_legend() # re-wrap lines if necessary
self.rescale()
def rescale(self):
if self.legend:
leg_height = self.legend.boundingRect().height()
leg_extra = 1.5
else:
leg_height = 0
leg_extra = 1
vw, vh = self.view.width(), self.view.height() - leg_height
scale = min(vw / (self.size_x + 1),
vh / ((self.size_y + leg_extra) * self._grid_factors[1]))
self.view.setTransform(QTransform.fromScale(scale, scale))
if self.hexagonal:
self.view.setSceneRect(0, -1, self.size_x - 1,
(self.size_y + leg_extra) * sqrt3_2 +
leg_height / scale)
else:
self.view.setSceneRect(-0.25, -0.25, self.size_x - 0.5,
self.size_y - 0.5 + leg_height / scale)
def update_output(self):
if self.data is None:
self.Outputs.selected_data.send(None)
self.Outputs.annotated_data.send(None)
return
indices = np.zeros(len(self.data), dtype=int)
if self.selection is not None and np.any(self.selection):
for y in range(self.size_y):
for x in range(self.size_x):
rows = self.get_member_indices(x, y)
indices[rows] = self.selection[x, y]
if np.any(indices):
sel_data = create_groups_table(self.data, indices, False, "Group")
self.Outputs.selected_data.send(sel_data)
else:
self.Outputs.selected_data.send(None)
if np.max(indices) > 1:
annotated = create_groups_table(self.data, indices)
else:
annotated = create_annotated_table(self.data,
np.flatnonzero(indices))
self.Outputs.annotated_data.send(annotated)
def set_color_bins(self):
if self.attr_color is None:
self.thresholds = self.bin_labels = self.colors = None
elif self.attr_color.is_discrete:
self.thresholds = self.bin_labels = None
self.colors = self.attr_color.palette
else:
col = self.data.get_column_view(self.attr_color)[0].astype(float)
if self.attr_color.is_time:
binning = time_binnings(col, min_bins=4)[-1]
else:
binning = decimal_binnings(col, min_bins=4)[-1]
self.thresholds = binning.thresholds[1:-1]
self.bin_labels = (binning.labels[1:-1],
binning.short_labels[1:-1])
palette = BinnedContinuousPalette.from_palette(
self.attr_color.palette, binning.thresholds)
self.colors = palette
def create_legend(self):
if self.legend is not None:
self.scene.removeItem(self.legend)
self.legend = None
if self.attr_color is None:
return
if self.attr_color.is_discrete:
names = self.attr_color.values
else:
names = self._bin_names()
items = []
size = 8
for name, color in zip(names, self.colors.qcolors):
item = QGraphicsItemGroup()
item.addToGroup(
CanvasRectangle(None, -size / 2, -size / 2, size, size,
Qt.gray, color))
item.addToGroup(CanvasText(None, name, size, 0, Qt.AlignVCenter))
items.append(item)
self.legend = wrap_legend_items(items,
hspacing=20,
vspacing=16 + size,
max_width=self.view.width() - 25)
self.legend.setFlags(self.legend.ItemIgnoresTransformations)
self.legend.setTransform(
QTransform.fromTranslate(-self.legend.boundingRect().width() / 2,
0))
self.scene.addItem(self.legend)
self.set_legend_pos()
def _bin_names(self):
labels, short_labels = self.bin_labels
return \
[f"< {labels[0]}"] \
+ [f"{x} - {y}" for x, y in zip(labels, short_labels[1:])] \
+ [f"≥ {labels[-1]}"]
def set_legend_pos(self):
if self.legend is None:
return
self.legend.setPos(self.size_x / 2,
(self.size_y + 0.2 + 0.3 * self.hexagonal) *
self._grid_factors[1])
def send_report(self):
self.report_plot()
if self.attr_color:
self.report_caption(
f"Self-organizing map colored by '{self.attr_color.name}'")
class OWBoxPlot(widget.OWWidget):
"""
Here's how the widget's functions call each other:
- `set_data` is a signal handler fills the list boxes and calls
`grouping_changed`.
- `grouping_changed` handles changes of grouping attribute: it enables or
disables the box for ordering, orders attributes and calls `attr_changed`.
- `attr_changed` handles changes of attribute. It recomputes box data by
calling `compute_box_data`, shows the appropriate display box
(discrete/continuous) and then calls`layout_changed`
- `layout_changed` constructs all the elements for the scene (as lists of
QGraphicsItemGroup) and calls `display_changed`. It is called when the
attribute or grouping is changed (by attr_changed) and on resize event.
- `display_changed` puts the elements corresponding to the current display
settings on the scene. It is called when the elements are reconstructed
(layout is changed due to selection of attributes or resize event), or
when the user changes display settings or colors.
For discrete attributes, the flow is a bit simpler: the elements are not
constructed in advance (by layout_changed). Instead, layout_changed and
display_changed call display_changed_disc that draws everything.
"""
name = "箱形图"
description = "在方框图中可视化特征值的分布"
icon = "icons/BoxPlot.svg"
priority = 100
keywords = ["whisker"]
class Inputs:
data = Input("Data", Orange.data.Table)
class Outputs:
selected_data = Output("所选数据", Orange.data.Table, default=True)
annotated_data = Output(ANNOTATED_DATA_SIGNAL_NAME, Orange.data.Table)
#: Comparison types for continuous variables
CompareNone, CompareMedians, CompareMeans = 0, 1, 2
settingsHandler = DomainContextHandler()
conditions = ContextSetting([])
attribute = ContextSetting(None)
order_by_importance = Setting(False)
group_var = ContextSetting(None)
show_annotations = Setting(True)
compare = Setting(CompareMeans)
stattest = Setting(0)
sig_threshold = Setting(0.05)
stretched = Setting(True)
show_labels = Setting(True)
sort_freqs = Setting(False)
auto_commit = Setting(True)
_sorting_criteria_attrs = {
CompareNone: "",
CompareMedians: "median",
CompareMeans: "mean"
}
_pen_axis_tick = QPen(Qt.white, 5)
_pen_axis = QPen(Qt.darkGray, 3)
_pen_median = QPen(QBrush(QColor(0xff, 0xff, 0x00)), 2)
_pen_paramet = QPen(QBrush(QColor(0x33, 0x00, 0xff)), 2)
_pen_dotted = QPen(QBrush(QColor(0x33, 0x00, 0xff)), 1)
_pen_dotted.setStyle(Qt.DotLine)
_post_line_pen = QPen(Qt.lightGray, 2)
_post_grp_pen = QPen(Qt.lightGray, 4)
for pen in (_pen_paramet, _pen_median, _pen_dotted, _pen_axis,
_pen_axis_tick, _post_line_pen, _post_grp_pen):
pen.setCosmetic(True)
pen.setCapStyle(Qt.RoundCap)
pen.setJoinStyle(Qt.RoundJoin)
_pen_axis_tick.setCapStyle(Qt.FlatCap)
_box_brush = QBrush(QColor(0x33, 0x88, 0xff, 0xc0))
_axis_font = QFont()
_axis_font.setPixelSize(12)
_label_font = QFont()
_label_font.setPixelSize(11)
_attr_brush = QBrush(QColor(0x33, 0x00, 0xff))
graph_name = "盒式布景"
def __init__(self):
super().__init__()
self.stats = []
self.dataset = None
self.posthoc_lines = []
self.label_txts = self.mean_labels = self.boxes = self.labels = \
self.label_txts_all = self.attr_labels = self.order = []
self.p = -1.0
self.scale_x = self.scene_min_x = self.scene_width = 0
self.label_width = 0
self.attrs = VariableListModel()
view = gui.listView(self.controlArea,
self,
"attribute",
box="变量",
model=self.attrs,
callback=self.attr_changed)
view.setMinimumSize(QSize(30, 30))
# Any other policy than Ignored will let the QListBox's scrollbar
# set the minimal height (see the penultimate paragraph of
# http://doc.qt.io/qt-4.8/qabstractscrollarea.html#addScrollBarWidget)
view.setSizePolicy(QSizePolicy.Expanding, QSizePolicy.Ignored)
gui.separator(view.box, 6, 6)
self.cb_order = gui.checkBox(view.box,
self,
"order_by_importance",
"按相关性排序",
tooltip="由𝜒²或方差对子群排序",
callback=self.apply_sorting)
self.group_vars = DomainModel(placeholder="None",
separators=False,
valid_types=Orange.data.DiscreteVariable)
self.group_view = view = gui.listView(self.controlArea,
self,
"group_var",
box="子群",
model=self.group_vars,
callback=self.grouping_changed)
view.setEnabled(False)
view.setMinimumSize(QSize(30, 30))
# See the comment above
view.setSizePolicy(QSizePolicy.Expanding, QSizePolicy.Ignored)
# TODO: move Compare median/mean to grouping box
# The vertical size policy is needed to let only the list views expand
self.display_box = gui.vBox(self.controlArea,
"Display",
sizePolicy=(QSizePolicy.Minimum,
QSizePolicy.Maximum),
addSpace=False)
gui.checkBox(self.display_box,
self,
"show_annotations",
"Annotate",
callback=self.display_changed)
self.compare_rb = gui.radioButtonsInBox(
self.display_box,
self,
'compare',
btnLabels=["无比较", "中位数比较", "均值比较"],
callback=self.layout_changed)
# The vertical size policy is needed to let only the list views expand
self.stretching_box = box = gui.vBox(self.controlArea,
box="显示",
sizePolicy=(QSizePolicy.Minimum,
QSizePolicy.Fixed))
self.stretching_box.sizeHint = self.display_box.sizeHint
gui.checkBox(box,
self,
'stretched',
"拉杆",
callback=self.display_changed)
gui.checkBox(box,
self,
'show_labels',
"显示框标签",
callback=self.display_changed)
self.sort_cb = gui.checkBox(box,
self,
'sort_freqs',
"按子组频率排序",
callback=self.display_changed)
gui.rubber(box)
gui.auto_commit(self.controlArea, self, "auto_commit", "选择发送", "自动发送")
gui.vBox(self.mainArea, addSpace=True)
self.box_scene = QGraphicsScene()
self.box_scene.selectionChanged.connect(self.commit)
self.box_view = QGraphicsView(self.box_scene)
self.box_view.setRenderHints(QPainter.Antialiasing
| QPainter.TextAntialiasing
| QPainter.SmoothPixmapTransform)
self.box_view.viewport().installEventFilter(self)
self.mainArea.layout().addWidget(self.box_view)
e = gui.hBox(self.mainArea, addSpace=False)
self.infot1 = gui.widgetLabel(e, "<center>没有测试结果。</center>")
self.mainArea.setMinimumWidth(600)
self.stats = self.dist = self.conts = []
self.is_continuous = False
self.update_display_box()
def sizeHint(self):
return QSize(100, 500) # Vertical size is regulated by mainArea
def eventFilter(self, obj, event):
if obj is self.box_view.viewport() and \
event.type() == QEvent.Resize:
self.layout_changed()
return super().eventFilter(obj, event)
def reset_attrs(self, domain):
self.attrs[:] = [
var for var in chain(domain.class_vars, domain.metas,
domain.attributes) if var.is_primitive()
]
# noinspection PyTypeChecker
@Inputs.data
def set_data(self, dataset):
if dataset is not None and (not bool(dataset)
or not len(dataset.domain) and
not any(var.is_primitive()
for var in dataset.domain.metas)):
dataset = None
self.closeContext()
self.dataset = dataset
self.dist = self.stats = self.conts = []
self.group_var = None
self.attribute = None
if dataset:
domain = dataset.domain
self.group_vars.set_domain(domain)
self.group_view.setEnabled(len(self.group_vars) > 1)
self.reset_attrs(domain)
self.select_default_variables(domain)
self.openContext(self.dataset)
self.grouping_changed()
else:
self.reset_all_data()
self.commit()
def select_default_variables(self, domain):
# visualize first non-class variable, group by class (if present)
if len(self.attrs) > len(domain.class_vars):
self.attribute = self.attrs[len(domain.class_vars)]
elif self.attrs:
self.attribute = self.attrs[0]
if domain.class_var and domain.class_var.is_discrete:
self.group_var = domain.class_var
else:
self.group_var = None # Reset to trigger selection via callback
def apply_sorting(self):
def compute_score(attr):
if attr is group_var:
return 3
if attr.is_continuous:
# One-way ANOVA
col = data.get_column_view(attr)[0].astype(float)
groups = (col[group_col == i] for i in range(n_groups))
groups = (col[~np.isnan(col)] for col in groups)
groups = [group for group in groups if len(group)]
p = f_oneway(*groups)[1] if len(groups) > 1 else 2
else:
# Chi-square with the given distribution into groups
# (see degrees of freedom in computation of the p-value)
if not attr.values or not group_var.values:
return 2
observed = np.array(
contingency.get_contingency(data, group_var, attr))
observed = observed[observed.sum(axis=1) != 0, :]
observed = observed[:, observed.sum(axis=0) != 0]
if min(observed.shape) < 2:
return 2
expected = \
np.outer(observed.sum(axis=1), observed.sum(axis=0)) / \
np.sum(observed)
p = chisquare(observed.ravel(),
f_exp=expected.ravel(),
ddof=n_groups - 1)[1]
if math.isnan(p):
return 2
return p
data = self.dataset
if data is None:
return
domain = data.domain
attribute = self.attribute
group_var = self.group_var
if self.order_by_importance and group_var is not None:
n_groups = len(group_var.values)
group_col = data.get_column_view(group_var)[0] if \
domain.has_continuous_attributes(
include_class=True, include_metas=True) else None
self.attrs.sort(key=compute_score)
else:
self.reset_attrs(domain)
self.attribute = attribute
def reset_all_data(self):
self.clear_scene()
self.infot1.setText("")
self.attrs.clear()
self.group_vars.set_domain(None)
self.group_view.setEnabled(False)
self.is_continuous = False
self.update_display_box()
def grouping_changed(self):
self.cb_order.setEnabled(self.group_var is not None)
self.apply_sorting()
self.attr_changed()
def select_box_items(self):
temp_cond = self.conditions.copy()
for box in self.box_scene.items():
if isinstance(box, FilterGraphicsRectItem):
box.setSelected(
box.filter.conditions in [c.conditions for c in temp_cond])
def attr_changed(self):
self.compute_box_data()
self.update_display_box()
self.layout_changed()
if self.is_continuous:
heights = 90 if self.show_annotations else 60
self.box_view.centerOn(self.scene_min_x + self.scene_width / 2,
-30 - len(self.stats) * heights / 2 + 45)
else:
self.box_view.centerOn(self.scene_width / 2,
-30 - len(self.boxes) * 40 / 2 + 45)
def compute_box_data(self):
attr = self.attribute
if not attr:
return
dataset = self.dataset
self.is_continuous = attr.is_continuous
if dataset is None or not self.is_continuous and not attr.values or \
self.group_var and not self.group_var.values:
self.stats = self.dist = self.conts = []
return
if self.group_var:
self.dist = []
self.conts = contingency.get_contingency(dataset, attr,
self.group_var)
if self.is_continuous:
stats, label_texts = [], []
for i, cont in enumerate(self.conts):
if np.sum(cont[1]):
stats.append(BoxData(cont, attr, i, self.group_var))
label_texts.append(self.group_var.values[i])
self.stats = stats
self.label_txts_all = label_texts
else:
self.label_txts_all = \
[v for v, c in zip(self.group_var.values, self.conts)
if np.sum(c) > 0]
else:
self.dist = distribution.get_distribution(dataset, attr)
self.conts = []
if self.is_continuous:
self.stats = [BoxData(self.dist, attr, None)]
self.label_txts_all = [""]
self.label_txts = [
txts for stat, txts in zip(self.stats, self.label_txts_all)
if stat.n > 0
]
self.stats = [stat for stat in self.stats if stat.n > 0]
def update_display_box(self):
if self.is_continuous:
self.stretching_box.hide()
self.display_box.show()
self.compare_rb.setEnabled(self.group_var is not None)
else:
self.stretching_box.show()
self.display_box.hide()
self.sort_cb.setEnabled(self.group_var is not None)
def clear_scene(self):
self.closeContext()
self.box_scene.clearSelection()
self.box_scene.clear()
self.box_view.viewport().update()
self.attr_labels = []
self.labels = []
self.boxes = []
self.mean_labels = []
self.posthoc_lines = []
self.openContext(self.dataset)
def layout_changed(self):
attr = self.attribute
if not attr:
return
self.clear_scene()
if self.dataset is None or len(self.conts) == len(self.dist) == 0:
return
if not self.is_continuous:
self.display_changed_disc()
return
self.mean_labels = [
self.mean_label(stat, attr, lab)
for stat, lab in zip(self.stats, self.label_txts)
]
self.draw_axis()
self.boxes = [self.box_group(stat) for stat in self.stats]
self.labels = [
self.label_group(stat, attr, mean_lab)
for stat, mean_lab in zip(self.stats, self.mean_labels)
]
self.attr_labels = [
QGraphicsSimpleTextItem(lab) for lab in self.label_txts
]
for it in chain(self.labels, self.attr_labels):
self.box_scene.addItem(it)
self.display_changed()
def display_changed(self):
if self.dataset is None:
return
if not self.is_continuous:
self.display_changed_disc()
return
self.order = list(range(len(self.stats)))
criterion = self._sorting_criteria_attrs[self.compare]
if criterion:
vals = [getattr(stat, criterion) for stat in self.stats]
overmax = max((val for val in vals if val is not None), default=0) \
+ 1
vals = [val if val is not None else overmax for val in vals]
self.order = sorted(self.order, key=vals.__getitem__)
heights = 90 if self.show_annotations else 60
for row, box_index in enumerate(self.order):
y = (-len(self.stats) + row) * heights + 10
for item in self.boxes[box_index]:
self.box_scene.addItem(item)
item.setY(y)
labels = self.labels[box_index]
if self.show_annotations:
labels.show()
labels.setY(y)
else:
labels.hide()
label = self.attr_labels[box_index]
label.setY(y - 15 - label.boundingRect().height())
if self.show_annotations:
label.hide()
else:
stat = self.stats[box_index]
if self.compare == OWBoxPlot.CompareMedians and \
stat.median is not None:
pos = stat.median + 5 / self.scale_x
elif self.compare == OWBoxPlot.CompareMeans or stat.q25 is None:
pos = stat.mean + 5 / self.scale_x
else:
pos = stat.q25
label.setX(pos * self.scale_x)
label.show()
r = QRectF(self.scene_min_x, -30 - len(self.stats) * heights,
self.scene_width,
len(self.stats) * heights + 90)
self.box_scene.setSceneRect(r)
self.compute_tests()
self.show_posthoc()
self.select_box_items()
def display_changed_disc(self):
assert not self.is_continuous
self.clear_scene()
self.attr_labels = [
QGraphicsSimpleTextItem(lab) for lab in self.label_txts_all
]
if not self.stretched:
if self.group_var:
self.labels = [
QGraphicsTextItem("{}".format(int(sum(cont))))
for cont in self.conts if np.sum(cont) > 0
]
else:
self.labels = [QGraphicsTextItem(str(int(sum(self.dist))))]
self.order = list(range(len(self.attr_labels)))
self.draw_axis_disc()
if self.group_var:
self.boxes = \
[self.strudel(cont, i) for i, cont in enumerate(self.conts)
if np.sum(cont) > 0]
self.conts = self.conts[np.sum(np.array(self.conts), axis=1) > 0]
if self.sort_freqs:
# pylint: disable=invalid-unary-operand-type
self.order = sorted(
self.order, key=(-np.sum(self.conts, axis=1)).__getitem__)
else:
self.boxes = [self.strudel(self.dist)]
for row, box_index in enumerate(self.order):
y = (-len(self.boxes) + row) * 40 + 10
box = self.boxes[box_index]
bars, labels = box[::2], box[1::2]
self.__draw_group_labels(y, box_index)
if not self.stretched:
self.__draw_row_counts(y, box_index)
if self.show_labels and self.attribute is not self.group_var:
self.__draw_bar_labels(y, bars, labels)
self.__draw_bars(y, bars)
self.box_scene.setSceneRect(-self.label_width - 5,
-30 - len(self.boxes) * 40,
self.scene_width,
len(self.boxes * 40) + 90)
self.infot1.setText("")
self.select_box_items()
def __draw_group_labels(self, y, row):
"""Draw group labels
Parameters
----------
y: int
vertical offset of bars
row: int
row index
"""
label = self.attr_labels[row]
b = label.boundingRect()
label.setPos(-b.width() - 10, y - b.height() / 2)
self.box_scene.addItem(label)
def __draw_row_counts(self, y, row):
"""Draw row counts
Parameters
----------
y: int
vertical offset of bars
row: int
row index
"""
assert not self.is_continuous
label = self.labels[row]
b = label.boundingRect()
if self.group_var:
right = self.scale_x * sum(self.conts[row])
else:
right = self.scale_x * sum(self.dist)
label.setPos(right + 10, y - b.height() / 2)
self.box_scene.addItem(label)
def __draw_bar_labels(self, y, bars, labels):
"""Draw bar labels
Parameters
----------
y: int
vertical offset of bars
bars: List[FilterGraphicsRectItem]
list of bars being drawn
labels: List[QGraphicsTextItem]
list of labels for corresponding bars
"""
label = bar_part = None
for text_item, bar_part in zip(labels, bars):
label = self.Label(text_item.toPlainText())
label.setPos(bar_part.boundingRect().x(),
y - label.boundingRect().height() - 8)
label.setMaxWidth(bar_part.boundingRect().width())
self.box_scene.addItem(label)
def __draw_bars(self, y, bars):
"""Draw bars
Parameters
----------
y: int
vertical offset of bars
bars: List[FilterGraphicsRectItem]
list of bars to draw
"""
for item in bars:
item.setPos(0, y)
self.box_scene.addItem(item)
# noinspection PyPep8Naming
def compute_tests(self):
# The t-test and ANOVA are implemented here since they efficiently use
# the widget-specific data in self.stats.
# The non-parametric tests can't do this, so we use statistics.tests
def stat_ttest():
d1, d2 = self.stats
if d1.n == 0 or d2.n == 0:
return np.nan, np.nan
pooled_var = d1.var / d1.n + d2.var / d2.n
df = pooled_var ** 2 / \
((d1.var / d1.n) ** 2 / (d1.n - 1) +
(d2.var / d2.n) ** 2 / (d2.n - 1))
if pooled_var == 0:
return np.nan, np.nan
t = abs(d1.mean - d2.mean) / math.sqrt(pooled_var)
p = 2 * (1 - scipy.special.stdtr(df, t))
return t, p
# TODO: Check this function
# noinspection PyPep8Naming
def stat_ANOVA():
if any(stat.n == 0 for stat in self.stats):
return np.nan, np.nan
n = sum(stat.n for stat in self.stats)
grand_avg = sum(stat.n * stat.mean for stat in self.stats) / n
var_between = sum(stat.n * (stat.mean - grand_avg)**2
for stat in self.stats)
df_between = len(self.stats) - 1
var_within = sum(stat.n * stat.var for stat in self.stats)
df_within = n - len(self.stats)
F = (var_between / df_between) / (var_within / df_within)
p = 1 - scipy.special.fdtr(df_between, df_within, F)
return F, p
if self.compare == OWBoxPlot.CompareNone or len(self.stats) < 2:
t = ""
elif any(s.n <= 1 for s in self.stats):
t = "At least one group has just one instance," \
"cannot compute significance"
elif len(self.stats) == 2:
if self.compare == OWBoxPlot.CompareMedians:
t = ""
# z, self.p = tests.wilcoxon_rank_sum(
# self.stats[0].dist, self.stats[1].dist)
# t = "Mann-Whitney's z: %.1f (p=%.3f)" % (z, self.p)
else:
t, self.p = stat_ttest()
t = "Student's t: %.3f (p=%.3f)" % (t, self.p)
else:
if self.compare == OWBoxPlot.CompareMedians:
t = ""
# U, self.p = -1, -1
# t = "Kruskal Wallis's U: %.1f (p=%.3f)" % (U, self.p)
else:
F, self.p = stat_ANOVA()
t = "ANOVA: %.3f (p=%.3f)" % (F, self.p)
self.infot1.setText("<center>%s</center>" % t)
def mean_label(self, stat, attr, val_name):
label = QGraphicsItemGroup()
t = QGraphicsSimpleTextItem(
"%.*f" % (attr.number_of_decimals + 1, stat.mean), label)
t.setFont(self._label_font)
bbox = t.boundingRect()
w2, h = bbox.width() / 2, bbox.height()
t.setPos(-w2, -h)
tpm = QGraphicsSimpleTextItem(
" \u00b1 " + "%.*f" % (attr.number_of_decimals + 1, stat.dev),
label)
tpm.setFont(self._label_font)
tpm.setPos(w2, -h)
if val_name:
vnm = QGraphicsSimpleTextItem(val_name + ": ", label)
vnm.setFont(self._label_font)
vnm.setBrush(self._attr_brush)
vb = vnm.boundingRect()
label.min_x = -w2 - vb.width()
vnm.setPos(label.min_x, -h)
else:
label.min_x = -w2
return label
def draw_axis(self):
"""Draw the horizontal axis and sets self.scale_x"""
misssing_stats = not self.stats
stats = self.stats or [BoxData(np.array([[0.], [1.]]), self.attribute)]
mean_labels = self.mean_labels or [
self.mean_label(stats[0], self.attribute, "")
]
bottom = min(stat.a_min for stat in stats)
top = max(stat.a_max for stat in stats)
first_val, step = compute_scale(bottom, top)
while bottom <= first_val:
first_val -= step
bottom = first_val
no_ticks = math.ceil((top - first_val) / step) + 1
top = max(top, first_val + no_ticks * step)
gbottom = min(bottom, min(stat.mean - stat.dev for stat in stats))
gtop = max(top, max(stat.mean + stat.dev for stat in stats))
bv = self.box_view
viewrect = bv.viewport().rect().adjusted(15, 15, -15, -30)
self.scale_x = scale_x = viewrect.width() / (gtop - gbottom)
# In principle we should repeat this until convergence since the new
# scaling is too conservative. (No chance am I doing this.)
mlb = min(stat.mean + mean_lab.min_x / scale_x
for stat, mean_lab in zip(stats, mean_labels))
if mlb < gbottom:
gbottom = mlb
self.scale_x = scale_x = viewrect.width() / (gtop - gbottom)
self.scene_min_x = gbottom * scale_x
self.scene_width = (gtop - gbottom) * scale_x
val = first_val
decimals = max(3, 4 - int(math.log10(step)))
while True:
l = self.box_scene.addLine(val * scale_x, -1, val * scale_x, 1,
self._pen_axis_tick)
l.setZValue(100)
t = self.box_scene.addSimpleText(
repr(round(val, decimals)) if not misssing_stats else "?",
self._axis_font)
t.setFlags(t.flags() | QGraphicsItem.ItemIgnoresTransformations)
r = t.boundingRect()
t.setPos(val * scale_x - r.width() / 2, 8)
if val >= top:
break
val += step
self.box_scene.addLine(bottom * scale_x - 4, 0, top * scale_x + 4, 0,
self._pen_axis)
def draw_axis_disc(self):
"""
Draw the horizontal axis and sets self.scale_x for discrete attributes
"""
assert not self.is_continuous
if self.stretched:
if not self.attr_labels:
return
step = steps = 10
else:
if self.group_var:
max_box = max(float(np.sum(dist)) for dist in self.conts)
else:
max_box = float(np.sum(self.dist))
if max_box == 0:
self.scale_x = 1
return
_, step = compute_scale(0, max_box)
step = int(step) if step > 1 else 1
steps = int(math.ceil(max_box / step))
max_box = step * steps
bv = self.box_view
viewrect = bv.viewport().rect().adjusted(15, 15, -15, -30)
self.scene_width = viewrect.width()
lab_width = max(lab.boundingRect().width() for lab in self.attr_labels)
lab_width = max(lab_width, 40)
lab_width = min(lab_width, self.scene_width / 3)
self.label_width = lab_width
right_offset = 0 # offset for the right label
if not self.stretched and self.labels:
if self.group_var:
rows = list(zip(self.conts, self.labels))
else:
rows = [(self.dist, self.labels[0])]
# available space left of the 'group labels'
available = self.scene_width - lab_width - 10
scale_x = (available - right_offset) / max_box
max_right = max(
sum(dist) * scale_x + 10 + lbl.boundingRect().width()
for dist, lbl in rows)
right_offset = max(0, max_right - max_box * scale_x)
self.scale_x = scale_x = \
(self.scene_width - lab_width - 10 - right_offset) / max_box
self.box_scene.addLine(0, 0, max_box * scale_x, 0, self._pen_axis)
for val in range(0, step * steps + 1, step):
l = self.box_scene.addLine(val * scale_x, -1, val * scale_x, 1,
self._pen_axis_tick)
l.setZValue(100)
t = self.box_scene.addSimpleText(str(val), self._axis_font)
t.setPos(val * scale_x - t.boundingRect().width() / 2, 8)
if self.stretched:
self.scale_x *= 100
def label_group(self, stat, attr, mean_lab):
def centered_text(val, pos):
t = QGraphicsSimpleTextItem(
"%.*f" % (attr.number_of_decimals + 1, val), labels)
t.setFont(self._label_font)
bbox = t.boundingRect()
t.setPos(pos - bbox.width() / 2, 22)
return t
def line(x, down=1):
QGraphicsLineItem(x, 12 * down, x, 20 * down, labels)
def move_label(label, frm, to):
label.setX(to)
to += t_box.width() / 2
path = QPainterPath()
path.lineTo(0, 4)
path.lineTo(to - frm, 4)
path.lineTo(to - frm, 8)
p = QGraphicsPathItem(path)
p.setPos(frm, 12)
labels.addToGroup(p)
labels = QGraphicsItemGroup()
labels.addToGroup(mean_lab)
m = stat.mean * self.scale_x
mean_lab.setPos(m, -22)
line(m, -1)
if stat.median is not None:
msc = stat.median * self.scale_x
med_t = centered_text(stat.median, msc)
med_box_width2 = med_t.boundingRect().width() / 2
line(msc)
if stat.q25 is not None:
x = stat.q25 * self.scale_x
t = centered_text(stat.q25, x)
t_box = t.boundingRect()
med_left = msc - med_box_width2
if x + t_box.width() / 2 >= med_left - 5:
move_label(t, x, med_left - t_box.width() - 5)
else:
line(x)
if stat.q75 is not None:
x = stat.q75 * self.scale_x
t = centered_text(stat.q75, x)
t_box = t.boundingRect()
med_right = msc + med_box_width2
if x - t_box.width() / 2 <= med_right + 5:
move_label(t, x, med_right + 5)
else:
line(x)
return labels
def box_group(self, stat, height=20):
def line(x0, y0, x1, y1, *args):
return QGraphicsLineItem(x0 * scale_x, y0, x1 * scale_x, y1, *args)
scale_x = self.scale_x
box = []
whisker1 = line(stat.a_min, -1.5, stat.a_min, 1.5)
whisker2 = line(stat.a_max, -1.5, stat.a_max, 1.5)
vert_line = line(stat.a_min, 0, stat.a_max, 0)
mean_line = line(stat.mean, -height / 3, stat.mean, height / 3)
for it in (whisker1, whisker2, mean_line):
it.setPen(self._pen_paramet)
vert_line.setPen(self._pen_dotted)
var_line = line(stat.mean - stat.dev, 0, stat.mean + stat.dev, 0)
var_line.setPen(self._pen_paramet)
box.extend([whisker1, whisker2, vert_line, mean_line, var_line])
if stat.q25 is not None and stat.q75 is not None:
mbox = FilterGraphicsRectItem(stat.conditions, stat.q25 * scale_x,
-height / 2,
(stat.q75 - stat.q25) * scale_x,
height)
mbox.setBrush(self._box_brush)
mbox.setPen(QPen(Qt.NoPen))
mbox.setZValue(-200)
box.append(mbox)
if stat.median is not None:
median_line = line(stat.median, -height / 2, stat.median,
height / 2)
median_line.setPen(self._pen_median)
median_line.setZValue(-150)
box.append(median_line)
return box
def strudel(self, dist, group_val_index=None):
attr = self.attribute
ss = np.sum(dist)
box = []
if ss < 1e-6:
cond = [FilterDiscrete(attr, None)]
if group_val_index is not None:
cond.append(FilterDiscrete(self.group_var, [group_val_index]))
box.append(FilterGraphicsRectItem(cond, 0, -10, 1, 10))
cum = 0
for i, v in enumerate(dist):
if v < 1e-6:
continue
if self.stretched:
v /= ss
v *= self.scale_x
cond = [FilterDiscrete(attr, [i])]
if group_val_index is not None:
cond.append(FilterDiscrete(self.group_var, [group_val_index]))
rect = FilterGraphicsRectItem(cond, cum + 1, -6, v - 2, 12)
rect.setBrush(QBrush(QColor(*attr.colors[i])))
rect.setPen(QPen(Qt.NoPen))
if self.stretched:
tooltip = "{}: {:.2f}%".format(attr.values[i],
100 * dist[i] / sum(dist))
else:
tooltip = "{}: {}".format(attr.values[i], int(dist[i]))
rect.setToolTip(tooltip)
text = QGraphicsTextItem(attr.values[i])
box.append(rect)
box.append(text)
cum += v
return box
def commit(self):
self.conditions = [
item.filter for item in self.box_scene.selectedItems()
if item.filter
]
selected, selection = None, []
if self.conditions:
selected = Values(self.conditions, conjunction=False)(self.dataset)
selection = np.in1d(self.dataset.ids,
selected.ids,
assume_unique=True).nonzero()[0]
self.Outputs.selected_data.send(selected)
self.Outputs.annotated_data.send(
create_annotated_table(self.dataset, selection))
def show_posthoc(self):
def line(y0, y1):
it = self.box_scene.addLine(x, y0, x, y1, self._post_line_pen)
it.setZValue(-100)
self.posthoc_lines.append(it)
while self.posthoc_lines:
self.box_scene.removeItem(self.posthoc_lines.pop())
if self.compare == OWBoxPlot.CompareNone or len(self.stats) < 2:
return
if self.compare == OWBoxPlot.CompareMedians:
crit_line = "median"
else:
crit_line = "mean"
xs = []
height = 90 if self.show_annotations else 60
y_up = -len(self.stats) * height + 10
for pos, box_index in enumerate(self.order):
stat = self.stats[box_index]
x = getattr(stat, crit_line)
if x is None:
continue
x *= self.scale_x
xs.append(x * self.scale_x)
by = y_up + pos * height
line(by + 12, 3)
line(by - 12, by - 25)
used_to = []
last_to = to = 0
for frm, frm_x in enumerate(xs[:-1]):
for to in range(frm + 1, len(xs)):
if xs[to] - frm_x > 1.5:
to -= 1
break
if to in (last_to, frm):
continue
for rowi, used in enumerate(used_to):
if used < frm:
used_to[rowi] = to
break
else:
rowi = len(used_to)
used_to.append(to)
y = -6 - rowi * 6
it = self.box_scene.addLine(frm_x - 2, y, xs[to] + 2, y,
self._post_grp_pen)
self.posthoc_lines.append(it)
last_to = to
def get_widget_name_extension(self):
return self.attribute.name if self.attribute else None
def send_report(self):
self.report_plot()
text = ""
if self.attribute:
text += "Box plot for attribute '{}' ".format(self.attribute.name)
if self.group_var:
text += "grouped by '{}'".format(self.group_var.name)
if text:
self.report_caption(text)
class Label(QGraphicsSimpleTextItem):
"""Boxplot Label with settable maxWidth"""
# Minimum width to display label text
MIN_LABEL_WIDTH = 25
# padding bellow the text
PADDING = 3
__max_width = None
def maxWidth(self):
return self.__max_width
def setMaxWidth(self, max_width):
self.__max_width = max_width
def paint(self, painter, option, widget):
"""Overrides QGraphicsSimpleTextItem.paint
If label text is too long, it is elided
to fit into the allowed region
"""
if self.__max_width is None:
width = option.rect.width()
else:
width = self.__max_width
if width < self.MIN_LABEL_WIDTH:
# if space is too narrow, no label
return
fm = painter.fontMetrics()
text = fm.elidedText(self.text(), Qt.ElideRight, width)
painter.drawText(
option.rect.x(),
option.rect.y() + self.boundingRect().height() - self.PADDING,
text)
def initStyleOptionForIndex(self, option: QStyleOptionHeader,
logicalIndex: int) -> None:
"""
Similar to initStyleOptionForIndex in Qt 6.0 with the difference that
`isSectionSelected` is not used, only `sectionIntersectsSelection`
is used (isSectionSelected will scan the entire model column/row
when the whole column/row is selected).
"""
hover = self.logicalIndexAt(self.mapFromGlobal(QCursor.pos()))
pressed = self.__pressed
if self.highlightSections():
is_selected = self.__sectionIntersectsSelection
else:
is_selected = lambda _: False
state = QStyle.State_None
if self.isEnabled():
state |= QStyle.State_Enabled
if self.window().isActiveWindow():
state |= QStyle.State_Active
if self.sectionsClickable():
if logicalIndex == hover:
state |= QStyle.State_MouseOver
if logicalIndex == pressed:
state |= QStyle.State_Sunken
if self.highlightSections():
if is_selected(logicalIndex):
state |= QStyle.State_On
if self.isSortIndicatorShown() and \
self.sortIndicatorSection() == logicalIndex:
option.sortIndicator = (
QStyleOptionHeader.SortDown if self.sortIndicatorOrder()
== Qt.AscendingOrder else QStyleOptionHeader.SortUp)
style = self.style()
model = self.model()
orientation = self.orientation()
textAlignment = model.headerData(logicalIndex, self.orientation(),
Qt.TextAlignmentRole)
defaultAlignment = self.defaultAlignment()
textAlignment = (textAlignment if isinstance(textAlignment, int) else
defaultAlignment)
option.section = logicalIndex
option.state = QStyle.State(int(option.state) | int(state))
option.textAlignment = Qt.Alignment(int(textAlignment))
option.iconAlignment = Qt.AlignVCenter
text = model.headerData(logicalIndex, self.orientation(),
Qt.DisplayRole)
text = str(text) if text is not None else ""
option.text = text
icon = model.headerData(logicalIndex, self.orientation(),
Qt.DecorationRole)
try:
option.icon = QIcon(icon)
except (TypeError, ValueError): # pragma: no cover
pass
margin = 2 * style.pixelMetric(QStyle.PM_HeaderMargin, None, self)
headerArrowAlignment = style.styleHint(QStyle.SH_Header_ArrowAlignment,
None, self)
isHeaderArrowOnTheSide = headerArrowAlignment & Qt.AlignVCenter
if self.isSortIndicatorShown() and \
self.sortIndicatorSection() == logicalIndex \
and isHeaderArrowOnTheSide:
margin += style.pixelMetric(QStyle.PM_HeaderMarkSize, None, self)
if not option.icon.isNull():
margin += style.pixelMetric(QStyle.PM_SmallIconSize, None, self)
margin += style.pixelMetric(QStyle.PM_HeaderMargin, None, self)
if self.textElideMode() != Qt.ElideNone:
elideMode = self.textElideMode()
if hasattr(option, 'textElideMode'): # Qt 6.0
option.textElideMode = elideMode # pragma: no cover
else:
option.text = option.fontMetrics.elidedText(
option.text, elideMode,
option.rect.width() - margin)
foregroundBrush = model.headerData(logicalIndex, orientation,
Qt.ForegroundRole)
try:
foregroundBrush = QBrush(foregroundBrush)
except (TypeError, ValueError):
pass
else:
option.palette.setBrush(QPalette.ButtonText, foregroundBrush)
backgroundBrush = model.headerData(logicalIndex, orientation,
Qt.BackgroundRole)
try:
backgroundBrush = QBrush(backgroundBrush)
except (TypeError, ValueError):
pass
else:
option.palette.setBrush(QPalette.Button, backgroundBrush)
option.palette.setBrush(QPalette.Window, backgroundBrush)
# the section position
visual = self.visualIndex(logicalIndex)
assert visual != -1
first = self.__isFirstVisibleSection(visual)
last = self.__isLastVisibleSection(visual)
if first and last:
option.position = QStyleOptionHeader.OnlyOneSection
elif first:
option.position = QStyleOptionHeader.Beginning
elif last:
option.position = QStyleOptionHeader.End
else:
option.position = QStyleOptionHeader.Middle
option.orientation = orientation
# the selected position (in QHeaderView this is always computed even if
# highlightSections is False).
if self.highlightSections():
previousSelected = is_selected(self.logicalIndex(visual - 1))
nextSelected = is_selected(self.logicalIndex(visual + 1))
else:
previousSelected = nextSelected = False
if previousSelected and nextSelected:
option.selectedPosition = QStyleOptionHeader.NextAndPreviousAreSelected
elif previousSelected:
option.selectedPosition = QStyleOptionHeader.PreviousIsSelected
elif nextSelected:
option.selectedPosition = QStyleOptionHeader.NextIsSelected
else:
option.selectedPosition = QStyleOptionHeader.NotAdjacent
def mark_problematic_reader():
self.reader_combo.setItemData(self.reader_combo.currentIndex(),
QBrush(Qt.red), Qt.ForegroundRole)
def _setup_plot(self):
def get_minmax(points):
minmax = [float('inf'), float('-inf'), float('inf'), float('-inf')]
for pp in points:
for p in pp:
minmax[0] = min(p[0], minmax[0])
minmax[1] = max(p[0], minmax[1])
minmax[2] = min(p[1], minmax[2])
minmax[3] = max(p[1], minmax[3])
return minmax
self.plot.clear()
points = self.ca
variables = self.selected_vars()
colors = colorpalette.ColorPaletteGenerator(len(variables))
p_axes = self._p_axes()
if points is None:
return
if len(variables) == 2:
row_points = self.ca.row_factors[:, p_axes]
col_points = self.ca.col_factors[:, p_axes]
points = [row_points, col_points]
else:
points = self.ca.row_factors[:, p_axes]
counts = [len(var.values) for var in variables]
range_indices = np.cumsum([0] + counts)
ranges = zip(range_indices, range_indices[1:])
points = [points[s:e] for s, e in ranges]
minmax = get_minmax(points)
margin = abs(minmax[0] - minmax[1])
margin = margin * 0.05 if margin > 1e-10 else 1
self.plot.setXRange(minmax[0] - margin, minmax[1] + margin)
margin = abs(minmax[2] - minmax[3])
margin = margin * 0.05 if margin > 1e-10 else 1
self.plot.setYRange(minmax[2] - margin, minmax[3] + margin)
for i, (v, points) in enumerate(zip(variables, points)):
color_outline = colors[i]
color_outline.setAlpha(200)
color = QColor(color_outline)
color.setAlpha(120)
item = ScatterPlotItem(
x=points[:, 0],
y=points[:, 1],
brush=QBrush(color),
pen=pg.mkPen(color_outline.darker(120), width=1.5),
size=np.full((points.shape[0], ), 10.1),
)
self.plot.addItem(item)
for name, point in zip(v.values, points):
item = pg.TextItem(name, anchor=(0.5, 0))
self.plot.addItem(item)
item.setPos(point[0], point[1])
inertia = self.ca.inertia_of_axis()
if np.sum(inertia) == 0:
inertia = 100 * inertia
else:
inertia = 100 * inertia / np.sum(inertia)
ax = self.plot.getAxis("bottom")
ax.setLabel("Component {} ({:.1f}%)".format(p_axes[0] + 1,
inertia[p_axes[0]]))
ax = self.plot.getAxis("left")
ax.setLabel("Component {} ({:.1f}%)".format(p_axes[1] + 1,
inertia[p_axes[1]]))
def paintEvent(self, event):
painter = QStylePainter(self)
rect = self._subControlRect(QStyle.SC_SliderGroove)
is_horizontal = self.orientation() == Qt.Horizontal
minpos, maxpos = self.minimumPosition(), self.maximumPosition()
span = rect.width() if is_horizontal else rect.height()
x1 = QStyle.sliderPositionFromValue(
self.minimum(), self.maximum(), minpos, span, self.invertedAppearance())
x2 = QStyle.sliderPositionFromValue(
self.minimum(), self.maximum(), maxpos, span, self.invertedAppearance())
# Background
painter.fillRect(rect, Qt.white)
# Highlight
painter.setOpacity(.7)
if is_horizontal:
painter.fillRect(x1, rect.y(), x2 - x1, rect.height(), Qt.yellow)
else:
painter.fillRect(rect.x(), x1, rect.width(), x2 - x1, Qt.yellow)
painter.setOpacity(1)
# Histogram
if self._pixmap:
painter.drawPixmap(rect, self._pixmap, self._pixmap.rect())
# Frame
painter.setPen(QPen(QBrush(Qt.darkGray), 2))
painter.drawRect(rect)
# Handles
painter.setPen(QPen(QBrush(self._HANDLE_COLOR), self._HANDLE_WIDTH))
painter.setOpacity(9)
if is_horizontal:
painter.drawLine(x1, rect.y(), x1, rect.y() + rect.height())
painter.drawLine(x2, rect.y(), x2, rect.y() + rect.height())
else:
painter.drawLine(rect.x(), x1, rect.x() + rect.width(), x1)
painter.drawLine(rect.x(), x2, rect.x() + rect.width(), x2)
painter.setOpacity(1)
if self._show_text:
painter.setFont(QFont('Monospace', 7, QFont.Bold))
strMin, strMax = self.formatValues(minpos, maxpos)
widthMin = painter.fontMetrics().width(strMin)
widthMax = painter.fontMetrics().width(strMax)
height = painter.fontMetrics().height()
is_enough_space = x2 - x1 > 3 + (max(widthMax, widthMin)
if is_horizontal else
(2 * height + self._HANDLE_WIDTH))
if is_enough_space:
if is_horizontal:
painter.drawText(x1 + 3, rect.y() + height, strMin)
painter.drawText(x2 - widthMax - 2,
rect.y() + rect.height() - 3, strMax)
else:
painter.drawText(rect.x() + 1, x1 + height, strMin)
painter.drawText(rect.x() + rect.width() - widthMax - 1, x2 - 2, strMax)
# Tooltip
if self._showControlTooltip\
and (not self._show_text or not is_enough_space):
# (show control-drag tooltip)
painter.setFont(QFont('Monospace', 10, QFont.Normal))
text = "Hold {} to move time interval" \
.format("Cmd" if sys.platform == "darwin"
else "Ctrl")
w = painter.fontMetrics().width(text)
h = painter.fontMetrics().height()
brush = QColor(224, 224, 224, 212)
pen = QPen(Qt.NoPen)
rect = QRect(4, 4, w + 8, h + 4)
painter.setBrush(brush)
painter.setPen(pen)
painter.drawRect(rect)
painter.setPen(Qt.black)
painter.drawText(8, 4 + h, text)
def __init__(self, *args):
QGraphicsEllipseItem.__init__(self, *args)
self.setRect(-3.5, -3.5, 7., 7.)
self.setPen(QPen(Qt.NoPen))
self.setBrush(QBrush(QColor("#9CACB4")))
self.__hover = False
def __init__(self, pen = QPen(Qt.black), brush = QBrush(Qt.white), xData = None, yData = None, tooltip = None):
OWCurve.__init__(self, xData, yData, tooltip=tooltip)
self.set_pen(pen)
self.set_brush(brush)
self._item = QGraphicsRectItem(self)
def __init__(self, pen = QPen(Qt.black), brush = QBrush(Qt.white), xData = [], yData = [], tooltip = None):
OWCurve.__init__(self, xData, yData, tooltip=tooltip)
self._data_polygon = self.polygon_from_data(xData, yData)
self._polygon_item = QGraphicsPolygonItem(self)
self.set_pen(pen)
self.set_brush(brush)
from AnyQt.QtGui import (QBrush, QPen, QColor, QPainter, QPainterPath,
QTransform)
from AnyQt.QtWidgets import (QGraphicsItem, QGraphicsEllipseItem,
QGraphicsRectItem, QGraphicsTextItem,
QGraphicsLineItem, QGraphicsScene, QGraphicsView,
QStyle, QSizePolicy, QFormLayout)
from AnyQt.QtCore import (Qt, QRectF, QSize, QPointF, QLineF, QTimer,
pyqtSignal, pyqtProperty)
from Orange.widgets import gui
from Orange.widgets.widget import OWWidget
from Orange.widgets.settings import Setting
DefDroppletBrush = QBrush(Qt.darkGray)
class GraphNode:
def __init__(self, *_, **kwargs):
self.edges = kwargs.get("edges", set())
def graph_edges(self):
return self.edges
def graph_add_edge(self, edge):
self.edges.add(edge)
def __iter__(self):
for edge in self.edges:
yield edge.node2
def graph_nodes(self, atype=1):
def update_rect(self):
if self.colors is None:
self.rect.setBrush(QBrush(Qt.white))
else:
self.rect.setBrush(QBrush(self.gradient))
def __init__(self, *args):
super().__init__(*args)
self.setAcceptHoverEvents(True)
self.setAcceptedMouseButtons(Qt.LeftButton)
self.setBrush(QBrush(Qt.gray))
self.setPen(Qt.white)
class OWBoxPlot(widget.OWWidget):
name = "Box Plot"
description = "Visualize the distribution of feature values in a box plot."
icon = "icons/BoxPlot.svg"
priority = 100
keywords = ["whisker"]
class Inputs:
data = Input("Data", Orange.data.Table)
class Outputs:
selected_data = Output("Selected Data",
Orange.data.Table,
default=True)
annotated_data = Output(ANNOTATED_DATA_SIGNAL_NAME, Orange.data.Table)
class Warning(widget.OWWidget.Warning):
no_vars = widget.Msg(
"Data contains no categorical or numeric variables")
buttons_area_orientation = None
#: Comparison types for continuous variables
CompareNone, CompareMedians, CompareMeans = 0, 1, 2
settingsHandler = DomainContextHandler()
# If this was a list, context handler would try to match its elements to
# variable names!
selection = ContextSetting((), schema_only=True)
attribute = ContextSetting(None)
order_by_importance = Setting(False)
order_grouping_by_importance = Setting(False)
group_var = ContextSetting(None)
show_annotations = Setting(True)
compare = Setting(CompareMeans)
stattest = Setting(0)
sig_threshold = Setting(0.05)
stretched = Setting(True)
show_labels = Setting(True)
sort_freqs = Setting(False)
_sorting_criteria_attrs = {
CompareNone: "",
CompareMedians: "median",
CompareMeans: "mean"
}
_pen_axis_tick = QPen(Qt.white, 5)
_pen_axis = QPen(Qt.darkGray, 3)
_pen_median = QPen(QBrush(QColor(0xff, 0xff, 0x00)), 2)
_pen_paramet = QPen(QBrush(QColor(0x33, 0x00, 0xff)), 2)
_pen_dotted = QPen(QBrush(QColor(0x33, 0x00, 0xff)), 1)
_pen_dotted.setStyle(Qt.DotLine)
_post_line_pen = QPen(Qt.lightGray, 2)
_post_grp_pen = QPen(Qt.lightGray, 4)
for pen in (_pen_paramet, _pen_median, _pen_dotted, _pen_axis,
_pen_axis_tick, _post_line_pen, _post_grp_pen):
pen.setCosmetic(True)
pen.setCapStyle(Qt.RoundCap)
pen.setJoinStyle(Qt.RoundJoin)
_pen_axis_tick.setCapStyle(Qt.FlatCap)
_box_brush = QBrush(QColor(0x33, 0x88, 0xff, 0xc0))
_attr_brush = QBrush(QColor(0x33, 0x00, 0xff))
graph_name = "box_scene"
def __init__(self):
super().__init__()
self._axis_font = QFont()
self._axis_font.setPixelSize(12)
self._label_font = QFont()
self._label_font.setPixelSize(11)
self.dataset = None
self.stats = []
self.dist = self.conts = None
self.posthoc_lines = []
self.label_txts = self.mean_labels = self.boxes = self.labels = \
self.label_txts_all = self.attr_labels = self.order = []
self.scale_x = 1
self.scene_min_x = self.scene_max_x = self.scene_width = 0
self.label_width = 0
self.attrs = VariableListModel()
sorted_model = SortProxyModel(sortRole=Qt.UserRole)
sorted_model.setSourceModel(self.attrs)
sorted_model.sort(0)
box = gui.vBox(self.controlArea, "Variable")
view = self.attr_list = ListViewSearch()
view.setModel(sorted_model)
view.setSelectionMode(view.SingleSelection)
view.selectionModel().selectionChanged.connect(self.attr_changed)
view.setMinimumSize(QSize(30, 30))
# Any other policy than Ignored will let the QListBox's scrollbar
# set the minimal height (see the penultimate paragraph of
# http://doc.qt.io/qt-4.8/qabstractscrollarea.html#addScrollBarWidget)
view.setSizePolicy(QSizePolicy.Expanding, QSizePolicy.Ignored)
box.layout().addWidget(view)
gui.checkBox(box,
self,
"order_by_importance",
"Order by relevance to subgroups",
tooltip="Order by 𝜒² or ANOVA over the subgroups",
callback=self.apply_attr_sorting)
self.group_vars = VariableListModel(placeholder="None")
sorted_model = SortProxyModel(sortRole=Qt.UserRole)
sorted_model.setSourceModel(self.group_vars)
sorted_model.sort(0)
box = gui.vBox(self.controlArea, "Subgroups")
view = self.group_list = ListViewSearch()
view.setModel(sorted_model)
view.selectionModel().selectionChanged.connect(self.grouping_changed)
view.setMinimumSize(QSize(30, 30))
# See the comment above
view.setSizePolicy(QSizePolicy.Expanding, QSizePolicy.Ignored)
box.layout().addWidget(view)
gui.checkBox(box,
self,
"order_grouping_by_importance",
"Order by relevance to variable",
tooltip="Order by 𝜒² or ANOVA over the variable values",
callback=self.apply_group_sorting)
# TODO: move Compare median/mean to grouping box
# The vertical size policy is needed to let only the list views expand
self.display_box = gui.vBox(self.controlArea,
"Display",
sizePolicy=(QSizePolicy.Minimum,
QSizePolicy.Maximum))
gui.checkBox(self.display_box,
self,
"show_annotations",
"Annotate",
callback=self.update_graph)
self.compare_rb = gui.radioButtonsInBox(
self.display_box,
self,
'compare',
btnLabels=["No comparison", "Compare medians", "Compare means"],
callback=self.update_graph)
# The vertical size policy is needed to let only the list views expand
self.stretching_box = box = gui.vBox(self.controlArea,
box="Display",
sizePolicy=(QSizePolicy.Minimum,
QSizePolicy.Fixed))
self.stretching_box.sizeHint = self.display_box.sizeHint
gui.checkBox(box,
self,
'stretched',
"Stretch bars",
callback=self.update_graph,
stateWhenDisabled=False)
gui.checkBox(box,
self,
'show_labels',
"Show box labels",
callback=self.update_graph)
self.sort_cb = gui.checkBox(box,
self,
'sort_freqs',
"Sort by subgroup frequencies",
callback=self.update_graph,
stateWhenDisabled=False)
gui.vBox(self.mainArea)
self.box_scene = QGraphicsScene(self)
self.box_scene.selectionChanged.connect(self.on_selection_changed)
self.box_view = QGraphicsView(self.box_scene)
self.box_view.setRenderHints(QPainter.Antialiasing
| QPainter.TextAntialiasing
| QPainter.SmoothPixmapTransform)
self.box_view.viewport().installEventFilter(self)
self.mainArea.layout().addWidget(self.box_view)
self.stat_test = ""
self.mainArea.setMinimumWidth(300)
self.update_box_visibilities()
def sizeHint(self):
return QSize(900, 500)
def eventFilter(self, obj, event):
if obj is self.box_view.viewport() and \
event.type() == QEvent.Resize:
self.update_graph()
return super().eventFilter(obj, event)
@property
def show_stretched(self):
return self.stretched and self.group_var is not self.attribute
def reset_attrs(self):
domain = self.dataset.domain
self.attrs[:] = [
var for var in chain(domain.class_vars, domain.metas,
domain.attributes)
if var.is_primitive() and not var.attributes.get("hidden", False)
]
def reset_groups(self):
domain = self.dataset.domain
self.group_vars[:] = [None] + [
var for var in chain(domain.class_vars, domain.metas,
domain.attributes)
if var.is_discrete and not var.attributes.get("hidden", False)
]
@Inputs.data
def set_data(self, dataset):
self.closeContext()
self._reset_all_data()
if dataset and not (len(dataset.domain.variables)
or any(var.is_primitive()
for var in dataset.domain.metas)):
self.Warning.no_vars()
dataset = None
self.dataset = dataset
if dataset:
self.reset_attrs()
self.reset_groups()
self._select_default_variables()
self.openContext(self.dataset)
self._set_list_view_selections()
self.compute_box_data()
self.apply_attr_sorting()
self.apply_group_sorting()
self.update_graph()
self.select_box_items()
self.update_box_visibilities()
self.commit()
def _reset_all_data(self):
self.clear_scene()
self.Warning.no_vars.clear()
self.stats = []
self.dist = self.conts = None
self.group_var = None
self.attribute = None
self.stat_test = ""
self.attrs[:] = []
self.group_vars[:] = [None]
self.selection = ()
def _select_default_variables(self):
# visualize first non-class variable, group by class (if present)
domain = self.dataset.domain
if len(self.attrs) > len(domain.class_vars):
self.attribute = self.attrs[len(domain.class_vars)]
elif self.attrs:
self.attribute = self.attrs[0]
if domain.class_var and domain.class_var.is_discrete:
self.group_var = domain.class_var
def _set_list_view_selections(self):
for view, var, callback in ((self.attr_list, self.attribute,
self.attr_changed),
(self.group_list, self.group_var,
self.grouping_changed)):
src_model = view.model().sourceModel()
if var not in src_model:
continue
sel_model = view.selectionModel()
sel_model.selectionChanged.disconnect(callback)
row = src_model.indexOf(var)
index = view.model().index(row, 0)
sel_model.select(index, sel_model.ClearAndSelect)
self._ensure_selection_visible(view)
sel_model.selectionChanged.connect(callback)
def apply_attr_sorting(self):
def compute_score(attr):
# This function and the one in apply_group_sorting are similar, but
# different in too many details, so they are kept as separate
# functions.
# If you discover a bug in this function, check the other one, too.
if attr is group_var:
return 3
if attr.is_continuous:
# One-way ANOVA
col = data.get_column_view(attr)[0].astype(float)
groups = (col[group_col == i] for i in range(n_groups))
groups = (col[~np.isnan(col)] for col in groups)
groups = [group for group in groups if len(group) > 1]
p = f_oneway(*groups)[1] if len(groups) > 1 else 2
else:
p = self._chi_square(group_var, attr)[1]
if math.isnan(p):
return 2
return p
data = self.dataset
if data is None:
return
domain = data.domain
group_var = self.group_var
if self.order_by_importance and group_var is not None:
n_groups = len(group_var.values)
group_col = data.get_column_view(group_var)[0] if \
domain.has_continuous_attributes(
include_class=True, include_metas=True) else None
self._sort_list(self.attrs, self.attr_list, compute_score)
else:
self._sort_list(self.attrs, self.attr_list, None)
def apply_group_sorting(self):
def compute_stat(group):
# This function and the one in apply_attr_sorting are similar, but
# different in too many details, so they are kept as separate
# functions.
# If you discover a bug in this function, check the other one, too.
if group is attr:
return 3
if group is None:
return -1
if attr.is_continuous:
groups = self._group_cols(data, group, attr_col)
groups = [group for group in groups if len(group) > 1]
p = f_oneway(*groups)[1] if len(groups) > 1 else 2
else:
p = self._chi_square(group, attr)[1]
if math.isnan(p):
return 2
return p
data = self.dataset
if data is None:
return
attr = self.attribute
if self.order_grouping_by_importance:
if attr.is_continuous:
attr_col = data.get_column_view(attr)[0].astype(float)
self._sort_list(self.group_vars, self.group_list, compute_stat)
else:
self._sort_list(self.group_vars, self.group_list, None)
def _sort_list(self, source_model, view, key=None):
if key is None:
c = count()
def key(_): # pylint: disable=function-redefined
return next(c)
for i, attr in enumerate(source_model):
source_model.setData(source_model.index(i), key(attr), Qt.UserRole)
self._ensure_selection_visible(view)
@staticmethod
def _ensure_selection_visible(view):
selection = view.selectedIndexes()
if len(selection) == 1:
view.scrollTo(selection[0])
def _chi_square(self, group_var, attr):
# Chi-square with the given distribution into groups
if not attr.values or not group_var.values:
return 0, 2, 0
observed = np.array(
contingency.get_contingency(self.dataset, group_var, attr))
observed = observed[observed.sum(axis=1) != 0, :]
observed = observed[:, observed.sum(axis=0) != 0]
if min(observed.shape) < 2:
return 0, 2, 0
return chi2_contingency(observed)[:3]
def grouping_changed(self, selected):
if not selected:
return # should never come here
self.group_var = selected.indexes()[0].data(gui.TableVariable)
self._variables_changed(self.apply_attr_sorting)
def attr_changed(self, selected):
if not selected:
return # should never come here
self.attribute = selected.indexes()[0].data(gui.TableVariable)
self._variables_changed(self.apply_group_sorting)
def _variables_changed(self, sorting):
self.selection = ()
self.compute_box_data()
sorting()
self.update_graph()
self.update_box_visibilities()
self.commit()
def update_graph(self):
pending_selection = self.selection
self.box_scene.selectionChanged.disconnect(self.on_selection_changed)
try: # not for exceptions, just to reconnect after all possible paths
self.clear_scene()
if self.dataset is None or self.attribute is None:
return
if self.attribute.is_continuous:
self._display_changed_cont()
else:
self._display_changed_disc()
self.selection = pending_selection
self.draw_stat()
self.select_box_items()
if self.attribute.is_continuous:
heights = 90 if self.show_annotations else 60
self.box_view.centerOn(
self.scene_min_x + self.scene_width / 2,
-30 - len(self.stats) * heights / 2 + 45)
else:
self.box_view.centerOn(self.scene_width / 2,
-30 - len(self.boxes) * 40 / 2 + 45)
finally:
self.box_scene.selectionChanged.connect(self.on_selection_changed)
def select_box_items(self):
selection = set(self.selection)
for box in self.box_scene.items():
if isinstance(box, FilterGraphicsRectItem):
box.setSelected(box.data_range in selection)
def _group_cols(self, data, group, attr):
if isinstance(attr, np.ndarray):
attr_col = attr
else:
attr_col = data.get_column_view(group)[0].astype(float)
group_col = data.get_column_view(group)[0].astype(float)
groups = [attr_col[group_col == i] for i in range(len(group.values))]
groups = [col[~np.isnan(col)] for col in groups]
return groups
def compute_box_data(self):
attr = self.attribute
if not attr:
return
dataset = self.dataset
if dataset is None \
or not attr.is_continuous and not attr.values \
or self.group_var and not self.group_var.values:
self.stats = []
self.dist = self.conts = None
return
if self.group_var:
self.dist = None
missing_val_str = f"missing '{self.group_var.name}'"
group_var_labels = self.group_var.values + ("", )
if self.attribute.is_continuous:
stats, label_texts = [], []
attr_col = dataset.get_column_view(attr)[0].astype(float)
for group, value in \
zip(self._group_cols(dataset, self.group_var, attr_col),
group_var_labels):
if group.size:
stats.append(BoxData(group, value))
label_texts.append(value or missing_val_str)
self.stats = stats
self.label_txts_all = label_texts
else:
self.conts = contingency.get_contingency(
dataset, attr, self.group_var)
self.label_txts_all = [
v or missing_val_str for v, c in zip(
group_var_labels, self.conts.array_with_unknowns)
if np.sum(c) > 0
]
else:
self.conts = None
if self.attribute.is_continuous:
attr_col = dataset.get_column_view(attr)[0].astype(float)
self.stats = [BoxData(attr_col)]
else:
self.dist = distribution.get_distribution(dataset, attr)
self.label_txts_all = [""]
self.label_txts = [
txts for stat, txts in zip(self.stats, self.label_txts_all)
if stat.n > 0
]
self.stats = [stat for stat in self.stats if stat.n > 0]
def update_box_visibilities(self):
self.controls.stretched.setDisabled(self.group_var is self.attribute)
if not self.attribute:
self.stretching_box.hide()
self.display_box.hide()
elif self.attribute.is_continuous:
self.stretching_box.hide()
self.display_box.show()
self.compare_rb.setEnabled(self.group_var is not None)
else:
self.stretching_box.show()
self.display_box.hide()
self.sort_cb.setEnabled(self.group_var is not None)
def clear_scene(self):
self.box_scene.clear()
self.box_view.viewport().update()
self.attr_labels = []
self.labels = []
self.boxes = []
self.mean_labels = []
self.posthoc_lines = []
def _display_changed_cont(self):
self.mean_labels = [
self.mean_label(stat, self.attribute, lab)
for stat, lab in zip(self.stats, self.label_txts)
]
self.draw_axis()
self.boxes = [self.box_group(stat) for stat in self.stats]
self.labels = [
self.label_group(stat, self.attribute, mean_lab)
for stat, mean_lab in zip(self.stats, self.mean_labels)
]
self.attr_labels = [
QGraphicsSimpleTextItem(lab) for lab in self.label_txts
]
for it in chain(self.labels, self.attr_labels):
self.box_scene.addItem(it)
self.order = list(range(len(self.stats)))
criterion = self._sorting_criteria_attrs[self.compare]
if criterion:
vals = [getattr(stat, criterion) for stat in self.stats]
overmax = max((val for val in vals if val is not None), default=0) \
+ 1
vals = [val if val is not None else overmax for val in vals]
self.order = sorted(self.order, key=vals.__getitem__)
heights = 90 if self.show_annotations else 60
for row, box_index in enumerate(self.order):
y = (-len(self.stats) + row) * heights + 10
for item in self.boxes[box_index]:
self.box_scene.addItem(item)
item.setY(y)
labels = self.labels[box_index]
if self.show_annotations:
labels.show()
labels.setY(y)
else:
labels.hide()
label = self.attr_labels[box_index]
label.setY(y - 15 - label.boundingRect().height())
if self.show_annotations:
label.hide()
else:
stat = self.stats[box_index]
if self.compare == OWBoxPlot.CompareMedians and \
stat.median is not None:
pos = stat.median + 5 / self.scale_x
elif self.compare == OWBoxPlot.CompareMeans or stat.q25 is None:
pos = stat.mean + 5 / self.scale_x
else:
pos = stat.q25
label.setX(pos * self.scale_x)
label.show()
r = QRectF(self.scene_min_x, -30 - len(self.stats) * heights,
self.scene_width,
len(self.stats) * heights + 90)
self.box_scene.setSceneRect(r)
self._compute_tests_cont()
self._show_posthoc()
def _display_changed_disc(self):
self.clear_scene()
self.attr_labels = [
QGraphicsSimpleTextItem(lab) for lab in self.label_txts_all
]
if not self.show_stretched:
if self.group_var:
self.labels = [
QGraphicsTextItem("{}".format(int(sum(cont))))
for cont in self.conts.array_with_unknowns
if np.sum(cont) > 0
]
else:
self.labels = [QGraphicsTextItem(str(int(sum(self.dist))))]
self.order = list(range(len(self.attr_labels)))
self.draw_axis_disc()
if self.group_var:
conts = self.conts.array_with_unknowns
self.boxes = [
self.strudel(cont, val)
for cont, val in zip(conts, self.group_var.values + ("", ))
if np.sum(cont) > 0
]
sums_ = np.sum(conts, axis=1)
sums_ = sums_[sums_ > 0] # only bars with sum > 0 are shown
if self.sort_freqs:
# pylint: disable=invalid-unary-operand-type
self.order = sorted(self.order, key=(-sums_).__getitem__)
else:
conts = self.dist.array_with_unknowns
self.boxes = [self.strudel(conts)]
sums_ = [np.sum(conts)]
for row, box_index in enumerate(self.order):
y = (-len(self.boxes) + row) * 40 + 10
box = self.boxes[box_index]
bars, labels = box[::2], box[1::2]
self.__draw_group_labels(y, box_index)
if not self.show_stretched:
self.__draw_row_counts(y, self.labels[box_index],
sums_[box_index])
if self.show_labels and self.attribute is not self.group_var:
self.__draw_bar_labels(y, bars, labels)
self.__draw_bars(y, bars)
self.box_scene.setSceneRect(-self.label_width - 5,
-30 - len(self.boxes) * 40,
self.scene_width,
len(self.boxes * 40) + 90)
self._compute_tests_disc()
def __draw_group_labels(self, y, row):
"""Draw group labels
Parameters
----------
y: int
vertical offset of bars
row: int
row index
"""
label = self.attr_labels[row]
b = label.boundingRect()
label.setPos(-b.width() - 10, y - b.height() / 2)
self.box_scene.addItem(label)
def __draw_row_counts(self, y, label, row_sum_):
"""Draw row counts
Parameters
----------
y: int
vertical offset of bars
label: QGraphicsSimpleTextItem
Label for group
row_sum_: int
Sum for the group
"""
assert not self.attribute.is_continuous
b = label.boundingRect()
right = self.scale_x * row_sum_
label.setPos(right + 10, y - b.height() / 2)
self.box_scene.addItem(label)
def __draw_bar_labels(self, y, bars, labels):
"""Draw bar labels
Parameters
----------
y: int
vertical offset of bars
bars: List[FilterGraphicsRectItem]
list of bars being drawn
labels: List[QGraphicsTextItem]
list of labels for corresponding bars
"""
for text_item, bar_part in zip(labels, bars):
label = self.Label(text_item.toPlainText())
label.setPos(bar_part.boundingRect().x(),
y - label.boundingRect().height() - 8)
label.setMaxWidth(bar_part.boundingRect().width())
self.box_scene.addItem(label)
def __draw_bars(self, y, bars):
"""Draw bars
Parameters
----------
y: int
vertical offset of bars
bars: List[FilterGraphicsRectItem]
list of bars to draw
"""
for item in bars:
item.setPos(0, y)
self.box_scene.addItem(item)
# noinspection PyPep8Naming
def _compute_tests_cont(self):
# The t-test and ANOVA are implemented here since they efficiently use
# the widget-specific data in self.stats.
# The non-parametric tests can't do this, so we use statistics.tests
# pylint: disable=comparison-with-itself
def stat_ttest():
d1, d2 = self.stats
if d1.n < 2 or d2.n < 2:
return np.nan, np.nan
pooled_var = d1.var / d1.n + d2.var / d2.n
# pylint: disable=comparison-with-itself
if pooled_var == 0 or np.isnan(pooled_var):
return np.nan, np.nan
df = pooled_var ** 2 / \
((d1.var / d1.n) ** 2 / (d1.n - 1) +
(d2.var / d2.n) ** 2 / (d2.n - 1))
t = abs(d1.mean - d2.mean) / math.sqrt(pooled_var)
p = 2 * (1 - scipy.special.stdtr(df, t))
return t, p
# TODO: Check this function
# noinspection PyPep8Naming
def stat_ANOVA():
if any(stat.n == 0 for stat in self.stats):
return np.nan, np.nan
n = sum(stat.n for stat in self.stats)
grand_avg = sum(stat.n * stat.mean for stat in self.stats) / n
var_between = sum(stat.n * (stat.mean - grand_avg)**2
for stat in self.stats)
df_between = len(self.stats) - 1
var_within = sum(stat.n * stat.var for stat in self.stats)
df_within = n - len(self.stats)
if var_within == 0 or df_within == 0 or df_between == 0:
return np.nan, np.nan
F = (var_between / df_between) / (var_within / df_within)
p = 1 - scipy.special.fdtr(df_between, df_within, F)
return F, p
n = len(self.dataset)
if self.compare == OWBoxPlot.CompareNone or len(self.stats) < 2:
t = ""
elif any(s.n <= 1 for s in self.stats):
t = "At least one group has just one instance, " \
"cannot compute significance"
elif len(self.stats) == 2:
if self.compare == OWBoxPlot.CompareMedians:
t = ""
# z, p = tests.wilcoxon_rank_sum(
# self.stats[0].dist, self.stats[1].dist)
# t = "Mann-Whitney's z: %.1f (p=%.3f)" % (z, p)
else:
t, p = stat_ttest()
t = "" if np.isnan(
t) else f"Student's t: {t:.3f} (p={p:.3f}, N={n})"
else:
if self.compare == OWBoxPlot.CompareMedians:
t = ""
# U, p = -1, -1
# t = "Kruskal Wallis's U: %.1f (p=%.3f)" % (U, p)
else:
F, p = stat_ANOVA()
t = "" if np.isnan(F) else f"ANOVA: {F:.3f} (p={p:.3f}, N={n})"
self.stat_test = t
def _compute_tests_disc(self):
if self.group_var is None or self.attribute is None:
self.stat_test = ""
else:
chi, p, dof = self._chi_square(self.group_var, self.attribute)
if np.isnan(p):
self.stat_test = ""
else:
self.stat_test = f"χ²: {chi:.2f} (p={p:.3f}, dof={dof})"
def mean_label(self, stat, attr, val_name):
label = QGraphicsItemGroup()
t = QGraphicsSimpleTextItem(attr.str_val(stat.mean), label)
t.setFont(self._label_font)
bbox = t.boundingRect()
w2, h = bbox.width() / 2, bbox.height()
t.setPos(-w2, -h)
tpm = QGraphicsSimpleTextItem(
" \u00b1 " + "%.*f" % (attr.number_of_decimals + 1, stat.dev),
label)
tpm.setFont(self._label_font)
tpm.setPos(w2, -h)
if val_name:
vnm = QGraphicsSimpleTextItem(val_name + ": ", label)
vnm.setFont(self._label_font)
vnm.setBrush(self._attr_brush)
vb = vnm.boundingRect()
label.min_x = -w2 - vb.width()
vnm.setPos(label.min_x, -h)
else:
label.min_x = -w2
return label
def draw_axis(self):
"""Draw the horizontal axis and sets self.scale_x"""
misssing_stats = not self.stats
stats = self.stats or [BoxData(np.array([0.]), self.attribute)]
mean_labels = self.mean_labels or [
self.mean_label(stats[0], self.attribute, "")
]
bottom = min(stat.a_min for stat in stats)
top = max(stat.a_max for stat in stats)
first_val, step = compute_scale(bottom, top)
while bottom <= first_val:
first_val -= step
bottom = first_val
no_ticks = math.ceil((top - first_val) / step) + 1
top = max(top, first_val + no_ticks * step)
gbottom = min(bottom, min(stat.mean - stat.dev for stat in stats))
gtop = max(top, max(stat.mean + stat.dev for stat in stats))
bv = self.box_view
viewrect = bv.viewport().rect().adjusted(15, 15, -15, -30)
self.scale_x = scale_x = viewrect.width() / (gtop - gbottom)
# In principle we should repeat this until convergence since the new
# scaling is too conservative. (No chance am I doing this.)
mlb = min(stat.mean + mean_lab.min_x / scale_x
for stat, mean_lab in zip(stats, mean_labels))
if mlb < gbottom:
gbottom = mlb
self.scale_x = scale_x = viewrect.width() / (gtop - gbottom)
self.scene_min_x = gbottom * scale_x
self.scene_max_x = gtop * scale_x
self.scene_width = self.scene_max_x - self.scene_min_x
val = first_val
last_text = self.scene_min_x
while True:
l = self.box_scene.addLine(val * scale_x, -1, val * scale_x, 1,
self._pen_axis_tick)
l.setZValue(100)
t = QGraphicsSimpleTextItem(
self.attribute.str_val(val) if not misssing_stats else "?")
t.setFont(self._axis_font)
t.setFlag(QGraphicsItem.ItemIgnoresTransformations)
r = t.boundingRect()
x_start = val * scale_x - r.width() / 2
x_finish = x_start + r.width()
if x_start > last_text + 10 and x_finish < self.scene_max_x:
t.setPos(x_start, 8)
self.box_scene.addItem(t)
last_text = x_finish
if val >= top:
break
val += step
self.box_scene.addLine(bottom * scale_x - 4, 0, top * scale_x + 4, 0,
self._pen_axis)
def draw_stat(self):
if self.stat_test:
label = QGraphicsSimpleTextItem(self.stat_test)
brect = self.box_scene.sceneRect()
label.setPos(brect.center().x() - label.boundingRect().width() / 2,
8 + self._axis_font.pixelSize() * 2)
label.setFlag(QGraphicsItem.ItemIgnoresTransformations)
self.box_scene.addItem(label)
def draw_axis_disc(self):
"""
Draw the horizontal axis and sets self.scale_x for discrete attributes
"""
assert not self.attribute.is_continuous
if self.show_stretched:
if not self.attr_labels:
return
step = steps = 10
else:
if self.group_var:
max_box = max(
float(np.sum(dist))
for dist in self.conts.array_with_unknowns)
else:
max_box = float(np.sum(self.dist.array_with_unknowns))
if max_box == 0:
self.scale_x = 1
return
_, step = compute_scale(0, max_box)
step = int(step) if step > 1 else 1
steps = int(math.ceil(max_box / step))
max_box = step * steps
bv = self.box_view
viewrect = bv.viewport().rect().adjusted(15, 15, -15, -30)
self.scene_width = viewrect.width()
lab_width = max(lab.boundingRect().width() for lab in self.attr_labels)
lab_width = max(lab_width, 40)
lab_width = min(lab_width, self.scene_width / 3)
self.label_width = lab_width
right_offset = 0 # offset for the right label
if not self.show_stretched and self.labels:
if self.group_var:
rows = list(zip(self.conts.array_with_unknowns, self.labels))
else:
rows = [(self.dist, self.labels[0])]
# available space left of the 'group labels'
available = self.scene_width - lab_width - 10
scale_x = (available - right_offset) / max_box
max_right = max(
sum(dist) * scale_x + 10 + lbl.boundingRect().width()
for dist, lbl in rows)
right_offset = max(0, max_right - max_box * scale_x)
self.scale_x = scale_x = \
(self.scene_width - lab_width - 10 - right_offset) / max_box
self.box_scene.addLine(0, 0, max_box * scale_x, 0, self._pen_axis)
for val in range(0, step * steps + 1, step):
l = self.box_scene.addLine(val * scale_x, -1, val * scale_x, 1,
self._pen_axis_tick)
l.setZValue(100)
t = self.box_scene.addSimpleText(str(val), self._axis_font)
t.setPos(val * scale_x - t.boundingRect().width() / 2, 8)
if self.show_stretched:
self.scale_x *= 100
def label_group(self, stat, attr, mean_lab):
def centered_text(val, pos):
t = QGraphicsSimpleTextItem(attr.str_val(val), labels)
t.setFont(self._label_font)
bbox = t.boundingRect()
t.setPos(pos - bbox.width() / 2, 22)
return t
def line(x, down=1):
QGraphicsLineItem(x, 12 * down, x, 20 * down, labels)
def move_label(label, frm, to):
label.setX(to)
to += t_box.width() / 2
path = QPainterPath()
path.lineTo(0, 4)
path.lineTo(to - frm, 4)
path.lineTo(to - frm, 8)
p = QGraphicsPathItem(path)
p.setPos(frm, 12)
labels.addToGroup(p)
labels = QGraphicsItemGroup()
labels.addToGroup(mean_lab)
m = stat.mean * self.scale_x
mean_lab.setPos(m, -22)
line(m, -1)
if stat.median is not None:
msc = stat.median * self.scale_x
med_t = centered_text(stat.median, msc)
med_box_width2 = med_t.boundingRect().width() / 2
line(msc)
if stat.q25 is not None:
x = stat.q25 * self.scale_x
t = centered_text(stat.q25, x)
t_box = t.boundingRect()
med_left = msc - med_box_width2
if x + t_box.width() / 2 >= med_left - 5:
move_label(t, x, med_left - t_box.width() - 5)
else:
line(x)
if stat.q75 is not None:
x = stat.q75 * self.scale_x
t = centered_text(stat.q75, x)
t_box = t.boundingRect()
med_right = msc + med_box_width2
if x - t_box.width() / 2 <= med_right + 5:
move_label(t, x, med_right + 5)
else:
line(x)
return labels
def box_group(self, stat, height=20):
def line(x0, y0, x1, y1, *args):
return QGraphicsLineItem(x0 * scale_x, y0, x1 * scale_x, y1, *args)
scale_x = self.scale_x
box = []
whisker1 = line(stat.a_min, -1.5, stat.a_min, 1.5)
whisker2 = line(stat.a_max, -1.5, stat.a_max, 1.5)
vert_line = line(stat.a_min, 0, stat.a_max, 0)
mean_line = line(stat.mean, -height / 3, stat.mean, height / 3)
for it in (whisker1, whisker2, mean_line):
it.setPen(self._pen_paramet)
vert_line.setPen(self._pen_dotted)
var_line = line(stat.mean - stat.dev, 0, stat.mean + stat.dev, 0)
var_line.setPen(self._pen_paramet)
box.extend([whisker1, whisker2, vert_line, mean_line, var_line])
if stat.q25 is not None or stat.q75 is not None:
# if any of them is None it means that its value is equal to median
box_from = stat.median if stat.q25 is None else stat.q25
box_to = stat.median if stat.q75 is None else stat.q75
mbox = FilterGraphicsRectItem(stat.data_range, box_from * scale_x,
-height / 2,
(box_to - box_from) * scale_x,
height)
mbox.setBrush(self._box_brush)
mbox.setPen(QPen(Qt.NoPen))
mbox.setZValue(-200)
box.append(mbox)
if stat.median is not None:
median_line = line(stat.median, -height / 2, stat.median,
height / 2)
median_line.setPen(self._pen_median)
median_line.setZValue(-150)
box.append(median_line)
return box
def strudel(self, dist, group_val=None):
attr = self.attribute
ss = np.sum(dist)
box = []
if ss < 1e-6:
cond = DiscDataRange(None, group_val)
box.append(FilterGraphicsRectItem(cond, 0, -10, 1, 10))
cum = 0
missing_val_str = f"missing '{attr.name}'"
values = attr.values + ("", )
colors = attr.palette.qcolors_w_nan
total = sum(dist)
for freq, value, color in zip(dist, values, colors):
if freq < 1e-6:
continue
v = freq
if self.show_stretched:
v /= ss
v *= self.scale_x
cond = DiscDataRange(value, group_val)
rect = FilterGraphicsRectItem(cond, cum + 1, -6, v - 2, 12)
rect.setBrush(QBrush(color))
rect.setPen(QPen(Qt.NoPen))
value = value or missing_val_str
if self.show_stretched:
tooltip = f"{value}: {100 * freq / total:.2f}%"
else:
tooltip = f"{value}: ({int(freq)})"
rect.setToolTip(tooltip)
text = QGraphicsTextItem(value)
box.append(rect)
box.append(text)
cum += v
return box
def on_selection_changed(self):
self.selection = tuple(item.data_range
for item in self.box_scene.selectedItems()
if item.data_range)
self.commit()
def commit(self):
conditions = self._gather_conditions()
if conditions:
selected = Values(conditions, conjunction=False)(self.dataset)
selection = np.in1d(self.dataset.ids,
selected.ids,
assume_unique=True).nonzero()[0]
else:
selected, selection = None, []
self.Outputs.selected_data.send(selected)
self.Outputs.annotated_data.send(
create_annotated_table(self.dataset, selection))
def _gather_conditions(self):
conditions = []
attr = self.attribute
group_attr = self.group_var
for data_range in self.selection:
if attr.is_discrete:
# If some value was removed from the data (in case settings are
# loaded from a scheme), do not include the corresponding
# filter; this is appropriate since data with such value does
# not exist anyway
if not data_range.value:
condition = IsDefined([attr], negate=True)
elif data_range.value not in attr.values:
continue
else:
condition = FilterDiscrete(attr, [data_range.value])
else:
condition = FilterContinuous(attr, FilterContinuous.Between,
data_range.low, data_range.high)
if data_range.group_value:
if not data_range.group_value:
grp_filter = IsDefined([group_attr], negate=True)
elif data_range.group_value not in group_attr.values:
continue
else:
grp_filter = FilterDiscrete(group_attr,
[data_range.group_value])
condition = Values([condition, grp_filter], conjunction=True)
conditions.append(condition)
return conditions
def _show_posthoc(self):
def line(y0, y1):
it = self.box_scene.addLine(x, y0, x, y1, self._post_line_pen)
it.setZValue(-100)
self.posthoc_lines.append(it)
while self.posthoc_lines:
self.box_scene.removeItem(self.posthoc_lines.pop())
if self.compare == OWBoxPlot.CompareNone or len(self.stats) < 2:
return
if self.compare == OWBoxPlot.CompareMedians:
crit_line = "median"
else:
crit_line = "mean"
xs = []
height = 90 if self.show_annotations else 60
y_up = -len(self.stats) * height + 10
for pos, box_index in enumerate(self.order):
stat = self.stats[box_index]
x = getattr(stat, crit_line)
if x is None:
continue
x *= self.scale_x
xs.append(x * self.scale_x)
by = y_up + pos * height
line(by + 12, 0)
used_to = []
last_to = to = 0
for frm, frm_x in enumerate(xs[:-1]):
for to in range(frm + 1, len(xs)):
if xs[to] - frm_x > 1.5:
to -= 1
break
if to in (last_to, frm):
continue
for rowi, used in enumerate(used_to):
if used < frm:
used_to[rowi] = to
break
else:
rowi = len(used_to)
used_to.append(to)
y = -6 - rowi * 6
it = self.box_scene.addLine(frm_x - 2, y, xs[to] + 2, y,
self._post_grp_pen)
self.posthoc_lines.append(it)
last_to = to
def get_widget_name_extension(self):
return self.attribute.name if self.attribute else None
def send_report(self):
self.report_plot()
text = ""
if self.attribute:
text += "Box plot for attribute '{}' ".format(self.attribute.name)
if self.group_var:
text += "grouped by '{}'".format(self.group_var.name)
if text:
self.report_caption(text)
class Label(QGraphicsSimpleTextItem):
"""Boxplot Label with settable maxWidth"""
# Minimum width to display label text
MIN_LABEL_WIDTH = 25
# padding bellow the text
PADDING = 3
__max_width = None
def maxWidth(self):
return self.__max_width
def setMaxWidth(self, max_width):
self.__max_width = max_width
def paint(self, painter, option, widget):
"""Overrides QGraphicsSimpleTextItem.paint
If label text is too long, it is elided
to fit into the allowed region
"""
if self.__max_width is None:
width = option.rect.width()
else:
width = self.__max_width
if width < self.MIN_LABEL_WIDTH:
# if space is too narrow, no label
return
fm = painter.fontMetrics()
text = fm.elidedText(self.text(), Qt.ElideRight, int(width))
painter.drawText(
int(option.rect.x()),
int(option.rect.y() + self.boundingRect().height() -
self.PADDING), text)
def hoverEnterEvent(self, event):
super().hoverEnterEvent(event)
self.setBrush(QBrush(QColor(100, 100, 100)))
self.update()
def text_format(foreground=Qt.black, weight=QFont.Normal):
fmt = QTextCharFormat()
fmt.setForeground(QBrush(foreground))
fmt.setFontWeight(weight)
return fmt
def hoverLeaveEvent(self, event):
super().hoverLeaveEvent(event)
self.setBrush(QBrush(QColor(200, 200, 200)))
self.update()
def __setup(self):
# Setup the subwidgets/groups/layout
smax = max((np.nanmax(g.scores) for g in self.__groups
if g.scores.size),
default=1)
smax = 1 if np.isnan(smax) else smax
smin = min((np.nanmin(g.scores) for g in self.__groups
if g.scores.size),
default=-1)
smin = -1 if np.isnan(smin) else smin
smin = min(smin, 0)
font = self.font()
font.setPixelSize(self.__barHeight)
axispen = QPen(Qt.black)
ax = pg.AxisItem(parent=self, orientation="top", maxTickLength=7,
pen=axispen)
ax.setRange(smin, smax)
self.__topScale = ax
layout = self.__layout
assert layout is self.layout()
layout.addItem(ax, 0, 2)
for i, group in enumerate(self.__groups):
silhouettegroup = BarPlotItem(parent=self)
silhouettegroup.setBrush(QBrush(QColor(*group.color)))
silhouettegroup.setPen(self.__pen)
silhouettegroup.setDataRange(smin, smax)
silhouettegroup.setPlotData(group.scores)
silhouettegroup.setPreferredBarSize(self.__barHeight)
silhouettegroup.setData(0, group.indices)
layout.addItem(silhouettegroup, i + 1, 2)
if group.label:
layout.addItem(Line(orientation=Qt.Vertical), i + 1, 1)
label = QGraphicsSimpleTextItem(
"{} ({})".format(group.label, len(group.scores)), self
)
label.setRotation(-90)
item = SimpleLayoutItem(
label,
anchor=(0., 1.0),
anchorItem=(0., 0.),
)
item.setSizePolicy(QSizePolicy.Fixed, QSizePolicy.Fixed)
layout.addItem(item, i + 1, 0, Qt.AlignCenter)
textlist = _SilhouettePlotTextListWidget(
self, font=font, elideMode=Qt.ElideRight,
alignment=Qt.AlignLeft | Qt.AlignVCenter
)
textlist.setMaximumWidth(750)
textlist.setFlag(TextListWidget.ItemClipsChildrenToShape, False)
sp = textlist.sizePolicy()
sp.setVerticalPolicy(QSizePolicy.Ignored)
textlist.setSizePolicy(sp)
if group.rownames is not None:
textlist.setItems(group.items)
textlist.setVisible(self.__rowNamesVisible)
else:
textlist.setVisible(False)
layout.addItem(textlist, i + 1, 3)
ax = pg.AxisItem(parent=self, orientation="bottom", maxTickLength=7,
pen=axispen)
ax.setRange(smin, smax)
self.__bottomScale = ax
layout.addItem(ax, len(self.__groups) + 1, 2)
def backgroundBrush(self):
brush = getattr(self, "_background_brush")
if brush is None:
brush = getattr(self.scene(), "defaultItemBrush", Qt.NoBrush)
return QBrush(brush)
def brush(self):
"""Return the items brush.
"""
return QBrush(self.__brush)
class OWDataSets(OWWidget):
name = "Datasets"
description = "Load a dataset from an online repository"
icon = "icons/DataSets.svg"
priority = 20
replaces = ["orangecontrib.prototypes.widgets.owdatasets.OWDataSets"]
keywords = ["online", "data sets"]
want_control_area = False
# The following constants can be overridden in a subclass
# to reuse this widget for a different repository
# Take care when refactoring! (used in e.g. single-cell)
INDEX_URL = "https://datasets.biolab.si/"
DATASET_DIR = "datasets"
# override HEADER_SCHEMA to define new columns
# if schema is changed override methods: self.assign_delegates and
# self.create_model
HEADER_SCHEMA = [['islocal', {
'label': ''
}], ['title', {
'label': 'Title'
}], ['size', {
'label': 'Size'
}], ['instances', {
'label': 'Instances'
}], ['variables', {
'label': 'Variables'
}], ['target', {
'label': 'Target'
}], ['tags', {
'label': 'Tags'
}]] # type: List[str, dict]
IndicatorBrushes = (QBrush(Qt.darkGray), QBrush(QColor(0, 192, 0)))
class Error(OWWidget.Error):
no_remote_datasets = Msg("Could not fetch dataset list")
class Warning(OWWidget.Warning):
only_local_datasets = Msg("Could not fetch datasets list, only local "
"cached datasets are shown")
class Outputs:
data = Output("Data", Orange.data.Table)
#: Selected dataset id
selected_id = settings.Setting(None) # type: Optional[str]
#: main area splitter state
splitter_state = settings.Setting(b'') # type: bytes
header_state = settings.Setting(b'') # type: bytes
def __init__(self):
super().__init__()
self.allinfo_local = {}
self.allinfo_remote = {}
self.local_cache_path = os.path.join(data_dir(), self.DATASET_DIR)
# current_output does not equal selected_id when, for instance, the
# data is still downloading
self.current_output = None
self._header_labels = [
header['label'] for _, header in self.HEADER_SCHEMA
]
self._header_index = namedtuple(
'_header_index', [info_tag for info_tag, _ in self.HEADER_SCHEMA])
self.Header = self._header_index(
*[index for index, _ in enumerate(self._header_labels)])
self.__awaiting_state = None # type: Optional[_FetchState]
self.filterLineEdit = QLineEdit(
textChanged=self.filter, placeholderText="Search for data set ...")
self.mainArea.layout().addWidget(self.filterLineEdit)
self.splitter = QSplitter(orientation=Qt.Vertical)
self.view = TreeViewWithReturn(
sortingEnabled=True,
selectionMode=QTreeView.SingleSelection,
alternatingRowColors=True,
rootIsDecorated=False,
editTriggers=QTreeView.NoEditTriggers,
uniformRowHeights=True,
toolTip="Press Return or double-click to send")
# the method doesn't exists yet, pylint: disable=unnecessary-lambda
self.view.doubleClicked.connect(self.commit)
self.view.returnPressed.connect(self.commit)
box = gui.widgetBox(self.splitter, "Description", addToLayout=False)
self.descriptionlabel = QLabel(
wordWrap=True,
textFormat=Qt.RichText,
)
self.descriptionlabel = QTextBrowser(
openExternalLinks=True,
textInteractionFlags=(Qt.TextSelectableByMouse
| Qt.LinksAccessibleByMouse))
self.descriptionlabel.setFrameStyle(QTextBrowser.NoFrame)
# no (white) text background
self.descriptionlabel.viewport().setAutoFillBackground(False)
box.layout().addWidget(self.descriptionlabel)
self.splitter.addWidget(self.view)
self.splitter.addWidget(box)
self.splitter.setSizes([300, 200])
self.splitter.splitterMoved.connect(lambda: setattr(
self, "splitter_state", bytes(self.splitter.saveState())))
self.mainArea.layout().addWidget(self.splitter)
proxy = QSortFilterProxyModel()
proxy.setFilterKeyColumn(-1)
proxy.setFilterCaseSensitivity(False)
self.view.setModel(proxy)
if self.splitter_state:
self.splitter.restoreState(self.splitter_state)
self.assign_delegates()
self.setBlocking(True)
self.setStatusMessage("Initializing")
self._executor = ThreadPoolExecutor(max_workers=1)
f = self._executor.submit(list_remote, self.INDEX_URL)
w = FutureWatcher(f, parent=self)
w.done.connect(self.__set_index)
def assign_delegates(self):
# NOTE: All columns must have size hinting delegates.
# QTreeView queries only the columns displayed in the viewport so
# the layout would be different depending in the horizontal scroll
# position
self.view.setItemDelegate(UniformHeightDelegate(self))
self.view.setItemDelegateForColumn(
self.Header.islocal,
UniformHeightIndicatorDelegate(self,
role=Qt.DisplayRole,
indicatorSize=4))
self.view.setItemDelegateForColumn(self.Header.size,
SizeDelegate(self))
self.view.setItemDelegateForColumn(self.Header.instances,
NumericalDelegate(self))
self.view.setItemDelegateForColumn(self.Header.variables,
NumericalDelegate(self))
self.view.resizeColumnToContents(self.Header.islocal)
def _parse_info(self, file_path):
if file_path in self.allinfo_remote:
info = self.allinfo_remote[file_path]
else:
info = self.allinfo_local[file_path]
islocal = file_path in self.allinfo_local
isremote = file_path in self.allinfo_remote
outdated = islocal and isremote and (
self.allinfo_remote[file_path].get('version', '') !=
self.allinfo_local[file_path].get('version', ''))
islocal &= not outdated
prefix = os.path.join('', *file_path[:-1])
filename = file_path[-1]
return Namespace(file_path=file_path,
prefix=prefix,
filename=filename,
islocal=islocal,
outdated=outdated,
**info)
def create_model(self):
allkeys = set(self.allinfo_local) | set(self.allinfo_remote)
allkeys = sorted(allkeys)
model = QStandardItemModel(self)
model.setHorizontalHeaderLabels(self._header_labels)
current_index = -1
for i, file_path in enumerate(allkeys):
datainfo = self._parse_info(file_path)
item1 = QStandardItem()
item1.setData(" " if datainfo.islocal else "", Qt.DisplayRole)
item1.setData(self.IndicatorBrushes[0], Qt.ForegroundRole)
item1.setData(datainfo, Qt.UserRole)
item2 = QStandardItem(datainfo.title)
item3 = QStandardItem()
item3.setData(datainfo.size, Qt.DisplayRole)
item4 = QStandardItem()
item4.setData(datainfo.instances, Qt.DisplayRole)
item5 = QStandardItem()
item5.setData(datainfo.variables, Qt.DisplayRole)
item6 = QStandardItem()
item6.setData(datainfo.target, Qt.DisplayRole)
if datainfo.target:
item6.setIcon(variable_icon(datainfo.target))
item7 = QStandardItem()
item7.setData(", ".join(datainfo.tags) if datainfo.tags else "",
Qt.DisplayRole)
row = [item1, item2, item3, item4, item5, item6, item7]
model.appendRow(row)
if os.path.join(*file_path) == self.selected_id:
current_index = i
return model, current_index
@Slot(object)
def __set_index(self, f):
# type: (Future) -> None
# set results from `list_remote` query.
assert QThread.currentThread() is self.thread()
assert f.done()
self.setBlocking(False)
self.setStatusMessage("")
self.allinfo_local = list_local(self.local_cache_path)
try:
self.allinfo_remote = f.result()
except Exception: # anytying can happen, pylint: disable=broad-except
log.exception("Error while fetching updated index")
if not self.allinfo_local:
self.Error.no_remote_datasets()
else:
self.Warning.only_local_datasets()
self.allinfo_remote = {}
model, current_index = self.create_model()
self.view.model().setSourceModel(model)
self.view.selectionModel().selectionChanged.connect(
self.__on_selection)
scw = self.view.setColumnWidth
width = self.view.fontMetrics().width
self.view.resizeColumnToContents(0)
scw(self.Header.title, width("X" * 37))
scw(self.Header.size,
20 + max(width("888 bytes "), width("9999.9 MB ")))
scw(self.Header.instances, 20 + width("100000000"))
scw(self.Header.variables, 20 + width("1000000"))
header = self.view.header()
header.restoreState(self.header_state)
if current_index != -1:
selmodel = self.view.selectionModel()
selmodel.select(
self.view.model().mapFromSource(model.index(current_index, 0)),
QItemSelectionModel.ClearAndSelect | QItemSelectionModel.Rows)
self.commit()
def __update_cached_state(self):
model = self.view.model().sourceModel()
localinfo = list_local(self.local_cache_path)
assert isinstance(model, QStandardItemModel)
allinfo = []
for i in range(model.rowCount()):
item = model.item(i, 0)
info = item.data(Qt.UserRole)
is_local = info.file_path in localinfo
is_current = (is_local and os.path.join(
self.local_cache_path, *info.file_path) == self.current_output)
item.setData(" " * (is_local + is_current), Qt.DisplayRole)
item.setData(self.IndicatorBrushes[is_current], Qt.ForegroundRole)
allinfo.append(info)
def selected_dataset(self):
"""
Return the current selected dataset info or None if not selected
Returns
-------
info : Optional[Namespace]
"""
rows = self.view.selectionModel().selectedRows(0)
assert 0 <= len(rows) <= 1
current = rows[0] if rows else None # type: Optional[QModelIndex]
if current is not None:
info = current.data(Qt.UserRole)
assert isinstance(info, Namespace)
else:
info = None
return info
def filter(self):
filter_string = self.filterLineEdit.text().strip()
proxyModel = self.view.model()
if proxyModel:
proxyModel.setFilterFixedString(filter_string)
def __on_selection(self):
# Main datasets view selection has changed
rows = self.view.selectionModel().selectedRows(0)
assert 0 <= len(rows) <= 1
current = rows[0] if rows else None # type: Optional[QModelIndex]
if current is not None:
current = self.view.model().mapToSource(current)
di = current.data(Qt.UserRole)
text = description_html(di)
self.descriptionlabel.setText(text)
self.selected_id = os.path.join(di.prefix, di.filename)
else:
self.descriptionlabel.setText("")
self.selected_id = None
def commit(self):
"""
Commit a dataset to the output immediately (if available locally) or
schedule download background and an eventual send.
During the download the widget is in blocking state
(OWWidget.isBlocking)
"""
di = self.selected_dataset()
if di is not None:
self.Error.clear()
if self.__awaiting_state is not None:
# disconnect from the __commit_complete
self.__awaiting_state.watcher.done.disconnect(
self.__commit_complete)
# .. and connect to update_cached_state
# self.__awaiting_state.watcher.done.connect(
# self.__update_cached_state)
# TODO: There are possible pending __progress_advance queued
self.__awaiting_state.pb.advance.disconnect(
self.__progress_advance)
self.progressBarFinished()
self.__awaiting_state = None
if not di.islocal:
pr = progress()
callback = lambda pr=pr: pr.advance.emit()
pr.advance.connect(self.__progress_advance,
Qt.QueuedConnection)
self.progressBarInit()
self.setStatusMessage("Fetching...")
self.setBlocking(True)
f = self._executor.submit(ensure_local,
self.INDEX_URL,
di.file_path,
self.local_cache_path,
force=di.outdated,
progress_advance=callback)
w = FutureWatcher(f, parent=self)
w.done.connect(self.__commit_complete)
self.__awaiting_state = _FetchState(f, w, pr)
else:
self.setStatusMessage("")
self.setBlocking(False)
self.commit_cached(di.file_path)
else:
self.load_and_output(None)
@Slot(object)
def __commit_complete(self, f):
# complete the commit operation after the required file has been
# downloaded
assert QThread.currentThread() is self.thread()
assert self.__awaiting_state is not None
assert self.__awaiting_state.future is f
if self.isBlocking():
self.progressBarFinished()
self.setBlocking(False)
self.setStatusMessage("")
self.__awaiting_state = None
try:
path = f.result()
# anything can happen here, pylint: disable=broad-except
except Exception as ex:
log.exception("Error:")
self.error(format_exception(ex))
path = None
self.load_and_output(path)
def commit_cached(self, file_path):
path = LocalFiles(self.local_cache_path).localpath(*file_path)
self.load_and_output(path)
@Slot()
def __progress_advance(self):
assert QThread.currentThread() is self.thread()
self.progressBarAdvance(1)
def onDeleteWidget(self):
super().onDeleteWidget()
if self.__awaiting_state is not None:
self.__awaiting_state.watcher.done.disconnect(
self.__commit_complete)
self.__awaiting_state.pb.advance.disconnect(
self.__progress_advance)
self.__awaiting_state = None
@staticmethod
def sizeHint():
return QSize(1100, 500)
def closeEvent(self, event):
self.splitter_state = bytes(self.splitter.saveState())
self.header_state = bytes(self.view.header().saveState())
super().closeEvent(event)
def load_and_output(self, path):
if path is None:
self.Outputs.data.send(None)
else:
data = self.load_data(path)
self.Outputs.data.send(data)
self.current_output = path
self.__update_cached_state()
@staticmethod
def load_data(path):
return Orange.data.Table(path)
def brush(self):
return QBrush(self.__brush)
def add_points():
nonlocal cur, image_token
if image_token != self._image_token:
return
batch = visible[cur:cur + self.N_POINTS_PER_ITER]
batch_lat = lat[batch]
batch_lon = lon[batch]
x, y = self.Projection.latlon_to_easting_northing(
batch_lat, batch_lon)
x, y = self.Projection.easting_northing_to_pixel(
x, y, zoom, origin, map_pane_pos)
if self._jittering:
dx, dy = self._jittering_offsets[batch].T
x, y = x + dx, y + dy
colors = (self._colorgen.getRGB(
self._scaled_color_values[batch]).tolist()
if self._color_attr else repeat((0xff, 0, 0)))
sizes = self._size_coef * \
(self._sizes[batch] if self._size_attr else np.tile(10, len(batch)))
opacity_subset, opacity_rest = self._opacity, int(.8 *
self._opacity)
for x, y, is_selected, size, color, _in_subset in \
zip(x, y, selected[batch], sizes, colors, in_subset[batch]):
pensize2, selpensize2 = (.35, 1.5) if size >= 5 else (.15, .7)
pensize2 *= self._size_coef
selpensize2 *= self._size_coef
size2 = size / 2
if is_selected:
painter.setPen(QPen(QBrush(Qt.green), 2 * selpensize2))
painter.drawEllipse(x - size2 - selpensize2,
y - size2 - selpensize2,
size + selpensize2, size + selpensize2)
color = QColor(*color)
if _in_subset:
color.setAlpha(opacity_subset)
painter.setBrush(QBrush(color))
painter.setPen(
QPen(QBrush(color.darker(180)), 2 * pensize2))
else:
color.setAlpha(opacity_rest)
painter.setBrush(Qt.NoBrush)
painter.setPen(
QPen(QBrush(color.lighter(120)), 2 * pensize2))
painter.drawEllipse(x - size2 - pensize2, y - size2 - pensize2,
size + pensize2, size + pensize2)
im.save(self._overlay_image_path, 'PNG')
self.evalJS('markersImageLayer.setUrl("{}#{}"); 0;'.format(
self.toFileURL(self._overlay_image_path), np.random.random()))
cur += self.N_POINTS_PER_ITER
if cur < len(visible):
QTimer.singleShot(10, add_points)
self._owwidget.progressBarAdvance(100 / n_iters, None)
else:
self._owwidget.progressBarFinished(None)
self._image_token = None
def color_for_cell(self, row, col):
return QBrush(QColor.fromHsv(120, self.colors[row, col], 255))
def __setup(self):
# Setup the subwidgets/groups/layout
smax = max(
(np.nanmax(g.scores) for g in self.__groups if g.scores.size),
default=1)
smax = 1 if np.isnan(smax) else smax
smin = min(
(np.nanmin(g.scores) for g in self.__groups if g.scores.size),
default=-1)
smin = -1 if np.isnan(smin) else smin
smin = min(smin, 0)
font = self.font()
font.setPixelSize(self.__barHeight)
axispen = QPen(Qt.black)
ax = pg.AxisItem(parent=self,
orientation="top",
maxTickLength=7,
pen=axispen)
ax.setRange(smin, smax)
self.layout().addItem(ax, 0, 2)
for i, group in enumerate(self.__groups):
silhouettegroup = BarPlotItem(parent=self)
silhouettegroup.setBrush(QBrush(QColor(*group.color)))
silhouettegroup.setPen(self.__pen)
silhouettegroup.setDataRange(smin, smax)
silhouettegroup.setPlotData(group.scores)
silhouettegroup.setPreferredBarSize(self.__barHeight)
silhouettegroup.setData(0, group.indices)
self.layout().addItem(silhouettegroup, i + 1, 2)
if group.label:
self.layout().addItem(Line(orientation=Qt.Vertical), i + 1, 1)
label = QGraphicsSimpleTextItem(self)
label.setText("{} ({})".format(escape(group.label),
len(group.scores)))
item = WrapperLayoutItem(label, Qt.Vertical, parent=self)
self.layout().addItem(item, i + 1, 0, Qt.AlignCenter)
textlist = TextListWidget(self, font=font)
sp = textlist.sizePolicy()
sp.setVerticalPolicy(QSizePolicy.Ignored)
textlist.setSizePolicy(sp)
textlist.setParent(self)
if group.rownames is not None:
textlist.setItems(group.items)
textlist.setVisible(self.__rowNamesVisible)
else:
textlist.setVisible(False)
self.layout().addItem(textlist, i + 1, 3)
ax = pg.AxisItem(parent=self,
orientation="bottom",
maxTickLength=7,
pen=axispen)
ax.setRange(smin, smax)
self.layout().addItem(ax, len(self.__groups) + 1, 2)
def display_contingency(self):
"""
Set the contingency to display.
"""
cont = self.contingencies
var, cvar = self.var, self.cvar
if cont is None or not len(cont):
return
self.plot.clear()
self.plot_prob.clear()
self._legend.clear()
self.tooltip_items = []
if self.show_prob:
self.ploti.showAxis('right')
else:
self.ploti.hideAxis('right')
bottomaxis = self.ploti.getAxis("bottom")
bottomaxis.setLabel(var.name)
bottomaxis.resizeEvent()
cvar_values = cvar.values
colors = [QColor(*col) for col in cvar.colors]
if var and var.is_continuous:
bottomaxis.setTicks(None)
weights, cols, cvar_values, curves = [], [], [], []
for i, dist in enumerate(cont):
v, W = dist
if len(v):
weights.append(numpy.sum(W))
cols.append(colors[i])
cvar_values.append(cvar.values[i])
curves.append(
ash_curve(dist,
cont,
m=OWDistributions.ASH_HIST,
smoothing_factor=self.smoothing_factor))
weights = numpy.array(weights)
sumw = numpy.sum(weights)
weights /= sumw
colors = cols
curves = [(X, Y * w) for (X, Y), w in zip(curves, weights)]
curvesline = [] #from histograms to lines
for X, Y in curves:
X = X + (X[1] - X[0]) / 2
X = X[:-1]
X = numpy.array(X)
Y = numpy.array(Y)
curvesline.append((X, Y))
for t in ["fill", "line"]:
curve_data = list(zip(curvesline, colors, weights,
cvar_values))
for (X, Y), color, w, cval in reversed(curve_data):
item = pg.PlotCurveItem()
pen = QPen(QBrush(color), 3)
pen.setCosmetic(True)
color = QColor(color)
color.setAlphaF(0.2)
item.setData(X,
Y / (w if self.relative_freq else 1),
antialias=True,
stepMode=False,
fillLevel=0 if t == "fill" else None,
brush=QBrush(color),
pen=pen)
self.plot.addItem(item)
if t == "line":
item.tooltip = "{}\n{}={}".format(
"Normalized density " if self.relative_freq else
"Density ", cvar.name, cval)
self.tooltip_items.append((self.plot, item))
if self.show_prob:
all_X = numpy.array(
numpy.unique(numpy.hstack([X for X, _ in curvesline])))
inter_X = numpy.array(
numpy.linspace(all_X[0], all_X[-1],
len(all_X) * 2))
curvesinterp = [
numpy.interp(inter_X, X, Y) for (X, Y) in curvesline
]
sumprob = numpy.sum(curvesinterp, axis=0)
legal = sumprob > 0.05 * numpy.max(sumprob)
i = len(curvesinterp) + 1
show_all = self.show_prob == i
for Y, color, cval in reversed(
list(zip(curvesinterp, colors, cvar_values))):
i -= 1
if show_all or self.show_prob == i:
item = pg.PlotCurveItem()
pen = QPen(QBrush(color), 3, style=Qt.DotLine)
pen.setCosmetic(True)
prob = Y[legal] / sumprob[legal]
item.setData(inter_X[legal],
prob,
antialias=True,
stepMode=False,
fillLevel=None,
brush=None,
pen=pen)
self.plot_prob.addItem(item)
item.tooltip = "Probability that \n" + cvar.name + "=" + cval
self.tooltip_items.append((self.plot_prob, item))
elif var and var.is_discrete:
bottomaxis.setTicks([list(enumerate(var.values))])
cont = numpy.array(cont)
maxh = 0 #maximal column height
maxrh = 0 #maximal relative column height
scvar = cont.sum(axis=1)
#a cvar with sum=0 with allways have distribution counts 0,
#therefore we can divide it by anything
scvar[scvar == 0] = 1
for i, (value, dist) in enumerate(zip(var.values, cont.T)):
maxh = max(maxh, max(dist))
maxrh = max(maxrh, max(dist / scvar))
for i, (value, dist) in enumerate(zip(var.values, cont.T)):
dsum = sum(dist)
geom = QRectF(i - 0.333, 0, 0.666,
maxrh if self.relative_freq else maxh)
if self.show_prob:
prob = dist / dsum
ci = 1.96 * numpy.sqrt(prob * (1 - prob) / dsum)
else:
ci = None
item = DistributionBarItem(
geom, dist / scvar /
maxrh if self.relative_freq else dist / maxh, colors)
self.plot.addItem(item)
tooltip = "\n".join("%s: %.*f" %
(n, 3 if self.relative_freq else 1, v)
for n, v in zip(
cvar_values, dist /
scvar if self.relative_freq else dist))
item.tooltip = "{} ({}={}):\n{}".format(
"Normalized frequency " if self.relative_freq else
"Frequency ", cvar.name, value, tooltip)
self.tooltip_items.append((self.plot, item))
if self.show_prob:
item.tooltip += "\n\nProbabilities:"
for ic, a in enumerate(dist):
if self.show_prob - 1 != ic and \
self.show_prob - 1 != len(dist):
continue
position = -0.333 + ((ic + 0.5) * 0.666 / len(dist))
if dsum < 1e-6:
continue
prob = a / dsum
if not 1e-6 < prob < 1 - 1e-6:
continue
ci = 1.96 * sqrt(prob * (1 - prob) / dsum)
item.tooltip += "\n%s: %.3f ± %.3f" % (cvar_values[ic],
prob, ci)
mark = pg.ScatterPlotItem()
errorbar = pg.ErrorBarItem()
pen = QPen(QBrush(QColor(0)), 1)
pen.setCosmetic(True)
errorbar.setData(x=[i + position],
y=[prob],
bottom=min(numpy.array([ci]), prob),
top=min(numpy.array([ci]), 1 - prob),
beam=numpy.array([0.05]),
brush=QColor(1),
pen=pen)
mark.setData([i + position], [prob],
antialias=True,
symbol="o",
fillLevel=None,
pxMode=True,
size=10,
brush=QColor(colors[ic]),
pen=pen)
self.plot_prob.addItem(errorbar)
self.plot_prob.addItem(mark)
for color, name in zip(colors, cvar_values):
self._legend.addItem(
ScatterPlotItem(pen=color, brush=color, size=10, shape="s"),
escape(name))
self._legend.show()
from orangecontrib.imageanalytics.widgets.owimageviewer import (ImageLoader,
Preview)
_log = logging.getLogger(__name__)
_ImageItem = namedtuple(
"_ImageItem",
[
"index", # Index in the input data table
"widget", # GraphicsThumbnailWidget displaying the image.
"url", # Composed final image url.
"future"
] # Future instance yielding an QImage
)
DEFAULT_SELECTION_BRUSH = QBrush(QColor(217, 232, 252, 192))
DEFAULT_SELECTION_PEN = QPen(QColor(125, 162, 206, 192))
class OWImageGrid(widget.OWWidget):
name = "Image Grid"
description = "Visualize images in a similarity grid"
icon = "icons/ImageGrid.svg"
priority = 160
keywords = ["image", "grid", "similarity"]
graph_name = "scene"
class Inputs:
data = Input("Embeddings", Orange.data.Table, default=True)
data_subset = Input("Data Subset", Orange.data.Table)
