class OWMDS(widget.OWWidget):
name = "MDS"
description = "Two-dimensional data projection by multidimensional " \
"scaling constructed from a distance matrix."
icon = "icons/MDS.svg"
inputs = [("Data", Orange.data.Table, "set_data", widget.Default),
("Distances", Orange.misc.DistMatrix, "set_disimilarity"),
("Data Subset", Orange.data.Table, "set_subset_data")]
outputs = [("Selected Data", Orange.data.Table, widget.Default),
("Data", Orange.data.Table)]
#: Initialization type
PCA, Random = 0, 1
#: Refresh rate
RefreshRate = [
("Every iteration", 1),
("Every 5 steps", 5),
("Every 10 steps", 10),
("Every 25 steps", 25),
("Every 50 steps", 50),
("None", -1)
]
JitterAmount = [
("None", 0),
("0.1 %", 0.1),
("0.5 %", 0.5),
("1 %", 1.0),
("2 %", 2.0)
]
#: Runtime state
Running, Finished, Waiting = 1, 2, 3
settingsHandler = settings.DomainContextHandler()
max_iter = settings.Setting(300)
initialization = settings.Setting(PCA)
refresh_rate = settings.Setting(3)
# output embedding role.
NoRole, AttrRole, AddAttrRole, MetaRole = 0, 1, 2, 3
output_embedding_role = settings.Setting(2)
autocommit = settings.Setting(True)
color_value = settings.ContextSetting("")
shape_value = settings.ContextSetting("")
size_value = settings.ContextSetting("")
label_value = settings.ContextSetting("")
label_only_selected = settings.Setting(False)
symbol_size = settings.Setting(8)
symbol_opacity = settings.Setting(230)
connected_pairs = settings.Setting(5)
jitter = settings.Setting(0)
legend_anchor = settings.Setting(((1, 0), (1, 0)))
graph_name = "plot.plotItem"
def __init__(self):
super().__init__()
self.matrix = None
self.data = None
self.subset_data = None # type: Optional[Orange.data.Table]
self.matrix_data = None
self.signal_data = None
self._pen_data = None
self._brush_data = None
self._shape_data = None
self._size_data = None
self._label_data = None
self._similar_pairs = None
self._scatter_item = None
self._legend_item = None
self._selection_mask = None
self._subset_mask = None # type: Optional[numpy.ndarray]
self._invalidated = False
self._effective_matrix = None
self.__update_loop = None
self.__state = OWMDS.Waiting
self.__in_next_step = False
self.__draw_similar_pairs = False
box = gui.vBox(self.controlArea, "MDS Optimization")
form = QtGui.QFormLayout(
labelAlignment=Qt.AlignLeft,
formAlignment=Qt.AlignLeft,
fieldGrowthPolicy=QtGui.QFormLayout.AllNonFixedFieldsGrow,
verticalSpacing=10
)
form.addRow("Max iterations:",
gui.spin(box, self, "max_iter", 10, 10 ** 4, step=1))
form.addRow("Initialization:",
gui.comboBox(box, self, "initialization",
items=["PCA (Torgerson)", "Random"],
callback=self.__invalidate_embedding))
box.layout().addLayout(form)
form.addRow("Refresh:",
gui.comboBox(
box, self, "refresh_rate",
items=[t for t, _ in OWMDS.RefreshRate],
callback=self.__invalidate_refresh))
gui.separator(box, 10)
self.runbutton = gui.button(
box, self, "Run", callback=self._toggle_run)
box = gui.vBox(self.controlArea, "Graph")
self.colorvar_model = itemmodels.VariableListModel()
common_options = dict(
sendSelectedValue=True, valueType=str, orientation=Qt.Horizontal,
labelWidth=50, contentsLength=12)
self.cb_color_value = gui.comboBox(
box, self, "color_value", label="Color:",
callback=self._on_color_index_changed, **common_options)
self.cb_color_value.setModel(self.colorvar_model)
self.shapevar_model = itemmodels.VariableListModel()
self.cb_shape_value = gui.comboBox(
box, self, "shape_value", label="Shape:",
callback=self._on_shape_index_changed, **common_options)
self.cb_shape_value.setModel(self.shapevar_model)
self.sizevar_model = itemmodels.VariableListModel()
self.cb_size_value = gui.comboBox(
box, self, "size_value", label="Size:",
callback=self._on_size_index_changed, **common_options)
self.cb_size_value.setModel(self.sizevar_model)
self.labelvar_model = itemmodels.VariableListModel()
self.cb_label_value = gui.comboBox(
box, self, "label_value", label="Label:",
callback=self._on_label_index_changed, **common_options)
self.cb_label_value.setModel(self.labelvar_model)
gui.checkBox(
gui.indentedBox(box), self, 'label_only_selected',
'Label only selected points', callback=self._on_label_index_changed)
form = QtGui.QFormLayout(
labelAlignment=Qt.AlignLeft,
formAlignment=Qt.AlignLeft,
fieldGrowthPolicy=QtGui.QFormLayout.AllNonFixedFieldsGrow,
verticalSpacing=10
)
form.addRow("Symbol size:",
gui.hSlider(box, self, "symbol_size",
minValue=1, maxValue=20,
callback=self._on_size_index_changed,
createLabel=False))
form.addRow("Symbol opacity:",
gui.hSlider(box, self, "symbol_opacity",
minValue=100, maxValue=255, step=100,
callback=self._on_color_index_changed,
createLabel=False))
form.addRow("Show similar pairs:",
gui.hSlider(
gui.hBox(self.controlArea),
self, "connected_pairs", minValue=0, maxValue=20,
createLabel=False,
callback=self._on_connected_changed))
form.addRow("Jitter:",
gui.comboBox(
box, self, "jitter",
items=[text for text, _ in self.JitterAmount],
callback=self._update_plot))
box.layout().addLayout(form)
gui.rubber(self.controlArea)
box = QtGui.QGroupBox("Zoom/Select", )
box.setLayout(QtGui.QHBoxLayout())
box.layout().setContentsMargins(2, 2, 2, 2)
group = QtGui.QActionGroup(self, exclusive=True)
def icon(name):
path = "icons/Dlg_{}.png".format(name)
path = pkg_resources.resource_filename(widget.__name__, path)
return QtGui.QIcon(path)
action_select = QtGui.QAction(
"Select", self, checkable=True, checked=True, icon=icon("arrow"),
shortcut=QtGui.QKeySequence(Qt.ControlModifier + Qt.Key_1))
action_zoom = QtGui.QAction(
"Zoom", self, checkable=True, checked=False, icon=icon("zoom"),
shortcut=QtGui.QKeySequence(Qt.ControlModifier + Qt.Key_2))
action_pan = QtGui.QAction(
"Pan", self, checkable=True, checked=False, icon=icon("pan_hand"),
shortcut=QtGui.QKeySequence(Qt.ControlModifier + Qt.Key_3))
action_reset_zoom = QtGui.QAction(
"Zoom to fit", self, icon=icon("zoom_reset"),
shortcut=QtGui.QKeySequence(Qt.ControlModifier + Qt.Key_0))
action_reset_zoom.triggered.connect(
lambda: self.plot.autoRange(padding=0.1,
items=[self._scatter_item]))
group.addAction(action_select)
group.addAction(action_zoom)
group.addAction(action_pan)
self.addActions(group.actions() + [action_reset_zoom])
action_select.setChecked(True)
def button(action):
b = QtGui.QToolButton()
b.setToolButtonStyle(Qt.ToolButtonIconOnly)
b.setDefaultAction(action)
return b
box.layout().addWidget(button(action_select))
box.layout().addWidget(button(action_zoom))
box.layout().addWidget(button(action_pan))
box.layout().addSpacing(4)
box.layout().addWidget(button(action_reset_zoom))
box.layout().addStretch()
self.controlArea.layout().addWidget(box)
box = gui.vBox(self.controlArea, "Output")
self.output_combo = gui.comboBox(
box, self, "output_embedding_role",
items=["Original features only",
"Coordinates only",
"Coordinates as features",
"Coordinates as meta attributes"],
callback=self._invalidate_output, addSpace=4)
gui.auto_commit(box, self, "autocommit", "Send Selected",
checkbox_label="Send selected automatically",
box=None)
self.plot = pg.PlotWidget(background="w", enableMenu=False)
self.plot.setAspectLocked(True)
self.plot.getPlotItem().hideAxis("bottom")
self.plot.getPlotItem().hideAxis("left")
self.plot.getPlotItem().hideButtons()
self.plot.setRenderHint(QtGui.QPainter.Antialiasing)
self.mainArea.layout().addWidget(self.plot)
self.selection_tool = PlotSelectionTool(parent=self)
self.zoom_tool = PlotZoomTool(parent=self)
self.pan_tool = PlotPanTool(parent=self)
self.pinch_tool = PlotPinchZoomTool(parent=self)
self.pinch_tool.setViewBox(self.plot.getViewBox())
self.selection_tool.setViewBox(self.plot.getViewBox())
self.selection_tool.selectionFinished.connect(self.__selection_end)
self.current_tool = self.selection_tool
def activate_tool(action):
self.current_tool.setViewBox(None)
if action is action_select:
active, cur = self.selection_tool, Qt.ArrowCursor
elif action is action_zoom:
active, cur = self.zoom_tool, Qt.ArrowCursor
elif action is action_pan:
active, cur = self.pan_tool, Qt.OpenHandCursor
self.current_tool = active
self.current_tool.setViewBox(self.plot.getViewBox())
self.plot.getViewBox().setCursor(QtGui.QCursor(cur))
group.triggered[QtGui.QAction].connect(activate_tool)
self._initialize()
@check_sql_input
def set_data(self, data):
"""Set the input data set.
Parameters
----------
data : Optional[Orange.data.Table]
"""
self.signal_data = data
if self.matrix is not None and data is not None and len(self.matrix) == len(data):
self.closeContext()
self.data = data
self.update_controls()
self.openContext(data)
else:
self._invalidated = True
self._selection_mask = None
def set_disimilarity(self, matrix):
"""Set the dissimilarity (distance) matrix.
Parameters
----------
matrix : Optional[Orange.misc.DistMatrix]
"""
self.matrix = matrix
if matrix is not None and matrix.row_items:
self.matrix_data = matrix.row_items
if matrix is None:
self.matrix_data = None
self._invalidated = True
self._selection_mask = None
def set_subset_data(self, subset_data):
"""Set a subset of `data` input to highlight in the plot.
Parameters
----------
subset_data: Optional[Orange.data.Table]
"""
self.subset_data = subset_data
# invalidate the pen/brush when the subset is changed
self._pen_data = self._brush_data = None
self._subset_mask = None # type: Optional[numpy.ndarray]
def _clear(self):
self._pen_data = None
self._brush_data = None
self._shape_data = None
self._size_data = None
self._label_data = None
self._similar_pairs = None
self.colorvar_model[:] = ["Same color"]
self.shapevar_model[:] = ["Same shape"]
self.sizevar_model[:] = ["Same size"]
self.labelvar_model[:] = ["No labels"]
self.color_value = self.colorvar_model[0]
self.shape_value = self.shapevar_model[0]
self.size_value = self.sizevar_model[0]
self.label_value = self.labelvar_model[0]
self.__set_update_loop(None)
self.__state = OWMDS.Waiting
def _clear_plot(self):
self.plot.clear()
self._scatter_item = None
if self._legend_item is not None:
anchor = legend_anchor_pos(self._legend_item)
if anchor is not None:
self.legend_anchor = anchor
if self._legend_item.scene() is not None:
self._legend_item.scene().removeItem(self._legend_item)
self._legend_item = None
def update_controls(self):
if self.data is None and getattr(self.matrix, 'axis', 1) == 0:
# Column-wise distances
attr = "Attribute names"
self.labelvar_model[:] = ["No labels", attr]
self.shapevar_model[:] = ["Same shape", attr]
self.colorvar_model[:] = ["Same solor", attr]
self.color_value = attr
self.shape_value = attr
else:
# initialize the graph state from data
domain = self.data.domain
all_vars = list(filter_visible(domain.variables + domain.metas))
cd_vars = [var for var in all_vars if var.is_primitive()]
disc_vars = [var for var in all_vars if var.is_discrete]
cont_vars = [var for var in all_vars if var.is_continuous]
shape_vars = [var for var in disc_vars
if len(var.values) <= len(ScatterPlotItem.Symbols) - 1]
self.colorvar_model[:] = chain(["Same color"],
[self.colorvar_model.Separator] if cd_vars else [],
cd_vars)
self.shapevar_model[:] = chain(["Same shape"],
[self.shapevar_model.Separator] if shape_vars else [],
shape_vars)
self.sizevar_model[:] = chain(["Same size", "Stress"],
[self.sizevar_model.Separator] if cont_vars else [],
cont_vars)
self.labelvar_model[:] = chain(["No labels"],
[self.labelvar_model.Separator] if all_vars else [],
all_vars)
if domain.class_var is not None:
self.color_value = domain.class_var.name
def _initialize(self):
# clear everything
self.closeContext()
self._clear()
self.data = None
self._effective_matrix = None
self.embedding = None
# if no data nor matrix is present reset plot
if self.signal_data is None and self.matrix is None:
return
if self.signal_data and self.matrix is not None and len(self.signal_data) != len(self.matrix):
self.error(1, "Data and distances dimensions do not match.")
self._update_plot()
return
self.error(1)
if self.signal_data:
self.data = self.signal_data
elif self.matrix_data:
self.data = self.matrix_data
if self.matrix is not None:
self._effective_matrix = self.matrix
if self.matrix.axis == 0 and self.data is self.matrix_data:
self.data = None
else:
preprocessed_data = Orange.projection.MDS().preprocess(self.data)
self._effective_matrix = Orange.distance.Euclidean(preprocessed_data)
self.update_controls()
self.openContext(self.data)
def _toggle_run(self):
if self.__state == OWMDS.Running:
self.stop()
self._invalidate_output()
else:
self.start()
def start(self):
if self.__state == OWMDS.Running:
return
elif self.__state == OWMDS.Finished:
# Resume/continue from a previous run
self.__start()
elif self.__state == OWMDS.Waiting and \
self._effective_matrix is not None:
self.__start()
def stop(self):
if self.__state == OWMDS.Running:
self.__set_update_loop(None)
def __start(self):
self.__draw_similar_pairs = False
X = self._effective_matrix
if self.embedding is not None:
init = self.embedding
elif self.initialization == OWMDS.PCA:
init = torgerson(X, n_components=2)
else:
init = None
# number of iterations per single GUI update step
_, step_size = OWMDS.RefreshRate[self.refresh_rate]
if step_size == -1:
step_size = self.max_iter
def update_loop(X, max_iter, step, init):
"""
return an iterator over successive improved MDS point embeddings.
"""
# NOTE: this code MUST NOT call into QApplication.processEvents
done = False
iterations_done = 0
oldstress = numpy.finfo(numpy.float).max
while not done:
step_iter = min(max_iter - iterations_done, step)
mds = Orange.projection.MDS(
dissimilarity="precomputed", n_components=2,
n_init=1, max_iter=step_iter)
mdsfit = mds.fit(X, init=init)
iterations_done += step_iter
embedding, stress = mdsfit.embedding_, mdsfit.stress_
stress /= numpy.sqrt(numpy.sum(embedding ** 2, axis=1)).sum()
if iterations_done >= max_iter:
done = True
elif (oldstress - stress) < mds.params["eps"]:
done = True
init = embedding
oldstress = stress
yield embedding, mdsfit.stress_, iterations_done / max_iter
self.__set_update_loop(update_loop(X, self.max_iter, step_size, init))
self.progressBarInit(processEvents=None)
def __set_update_loop(self, loop):
"""
Set the update `loop` coroutine.
The `loop` is a generator yielding `(embedding, stress, progress)`
tuples where `embedding` is a `(N, 2) ndarray` of current updated
MDS points, `stress` is the current stress and `progress` a float
ratio (0 <= progress <= 1)
If an existing update loop is already in palace it is interrupted
(closed).
.. note::
The `loop` must not explicitly yield control flow to the event
loop (i.e. call `QApplication.processEvents`)
"""
if self.__update_loop is not None:
self.__update_loop.close()
self.__update_loop = None
self.progressBarFinished(processEvents=None)
self.__update_loop = loop
if loop is not None:
self.progressBarInit(processEvents=None)
self.setStatusMessage("Running")
self.runbutton.setText("Stop")
self.__state = OWMDS.Running
QtGui.QApplication.postEvent(self, QEvent(QEvent.User))
else:
self.setStatusMessage("")
self.runbutton.setText("Start")
self.__state = OWMDS.Finished
def __next_step(self):
if self.__update_loop is None:
return
loop = self.__update_loop
try:
embedding, stress, progress = next(self.__update_loop)
assert self.__update_loop is loop
except StopIteration:
self.__set_update_loop(None)
self.unconditional_commit()
self.__draw_similar_pairs = True
self._update_plot()
self.plot.autoRange(padding=0.1, items=[self._scatter_item])
else:
self.progressBarSet(100.0 * progress, processEvents=None)
self.embedding = embedding
self._update_plot()
self.plot.autoRange(padding=0.1, items=[self._scatter_item])
# schedule next update
QtGui.QApplication.postEvent(
self, QEvent(QEvent.User), Qt.LowEventPriority)
def customEvent(self, event):
if event.type() == QEvent.User and self.__update_loop is not None:
if not self.__in_next_step:
self.__in_next_step = True
try:
self.__next_step()
finally:
self.__in_next_step = False
else:
warnings.warn(
"Re-entry in update loop detected. "
"A rogue `proccessEvents` is on the loose.",
RuntimeWarning)
# re-schedule the update iteration.
QtGui.QApplication.postEvent(self, QEvent(QEvent.User))
return super().customEvent(event)
def __invalidate_embedding(self):
# reset/invalidate the MDS embedding, to the default initialization
# (Random or PCA), restarting the optimization if necessary.
if self.embedding is None:
return
state = self.__state
if self.__update_loop is not None:
self.__set_update_loop(None)
X = self._effective_matrix
if self.initialization == OWMDS.PCA:
self.embedding = torgerson(X)
else:
self.embedding = numpy.random.rand(len(X), 2)
self._update_plot()
self.plot.autoRange(padding=0.1, items=[self._scatter_item])
# restart the optimization if it was interrupted.
if state == OWMDS.Running:
self.__start()
def __invalidate_refresh(self):
state = self.__state
if self.__update_loop is not None:
self.__set_update_loop(None)
# restart the optimization if it was interrupted.
# TODO: decrease the max iteration count by the already
# completed iterations count.
if state == OWMDS.Running:
self.__start()
def handleNewSignals(self):
if self._invalidated:
self._invalidated = False
self._initialize()
self.start()
self.__draw_similar_pairs = False
if self._subset_mask is None and self.subset_data is not None and \
self.data is not None:
self._subset_mask = numpy.in1d(self.data.ids, self.subset_data.ids)
self._update_plot()
self.plot.autoRange(padding=0.1)
self.unconditional_commit()
def _invalidate_output(self):
self.commit()
def _on_color_index_changed(self):
self._pen_data = None
self._update_plot()
def _on_shape_index_changed(self):
self._shape_data = None
self._update_plot()
def _on_size_index_changed(self):
self._size_data = None
self._update_plot()
def _on_label_index_changed(self):
self._label_data = None
self._update_plot()
def _on_connected_changed(self):
self._similar_pairs = None
self._update_plot()
def _update_plot(self):
self._clear_plot()
if self.embedding is not None:
self._setup_plot()
def _setup_plot(self):
have_data = self.data is not None
have_matrix_transposed = self.matrix is not None and not self.matrix.axis
plotstyle = mdsplotutils.plotstyle
size = self._effective_matrix.shape[0]
def column(data, variable):
a, _ = data.get_column_view(variable)
return a.ravel()
def attributes(matrix):
return matrix.row_items.domain.attributes
def scale(a):
dmin, dmax = numpy.nanmin(a), numpy.nanmax(a)
if dmax - dmin > 0:
return (a - dmin) / (dmax - dmin)
else:
return numpy.zeros_like(a)
def jitter(x, factor=1, rstate=None):
if rstate is None:
rstate = numpy.random.RandomState()
elif not isinstance(rstate, numpy.random.RandomState):
rstate = numpy.random.RandomState(rstate)
span = numpy.nanmax(x) - numpy.nanmin(x)
if span < numpy.finfo(x.dtype).eps * 100:
span = 1
a = factor * span / 100.
return x + (rstate.random_sample(x.shape) - 0.5) * a
if self._pen_data is None:
if self._selection_mask is not None:
pointflags = numpy.where(
self._selection_mask,
mdsplotutils.Selected, mdsplotutils.NoFlags)
else:
pointflags = None
color_index = self.cb_color_value.currentIndex()
if have_data and color_index > 0:
color_var = self.colorvar_model[color_index]
if color_var.is_discrete:
palette = colorpalette.ColorPaletteGenerator(
len(color_var.values)
)
plotstyle = plotstyle.updated(discrete_palette=palette)
else:
palette = None
color_data = mdsplotutils.color_data(
self.data, color_var, plotstyle=plotstyle)
color_data = numpy.hstack(
(color_data,
numpy.full((len(color_data), 1), self.symbol_opacity,
dtype=float))
)
pen_data = mdsplotutils.pen_data(color_data * 0.8, pointflags)
brush_data = mdsplotutils.brush_data(color_data)
elif have_matrix_transposed and \
self.colorvar_model[color_index] == 'Attribute names':
attr = attributes(self.matrix)
palette = colorpalette.ColorPaletteGenerator(len(attr))
color_data = [palette.getRGB(i) for i in range(len(attr))]
color_data = numpy.hstack((
color_data,
numpy.full((len(color_data), 1), self.symbol_opacity,
dtype=float))
)
pen_data = mdsplotutils.pen_data(color_data * 0.8, pointflags)
brush_data = mdsplotutils.brush_data(color_data)
else:
pen_data = make_pen(QtGui.QColor(Qt.darkGray), cosmetic=True)
if self._selection_mask is not None:
pen_data = numpy.array(
[pen_data, plotstyle.selected_pen])
pen_data = pen_data[self._selection_mask.astype(int)]
else:
pen_data = numpy.full(self._effective_matrix.dim, pen_data,
dtype=object)
brush_data = numpy.full(
size, pg.mkColor((192, 192, 192, self.symbol_opacity)),
dtype=object)
if self._subset_mask is not None and have_data and \
self._subset_mask.shape == (size, ):
# clear brush fill for non subset data
brush_data[~self._subset_mask] = QtGui.QBrush(Qt.NoBrush)
self._pen_data = pen_data
self._brush_data = brush_data
if self._shape_data is None:
shape_index = self.cb_shape_value.currentIndex()
if have_data and shape_index > 0:
Symbols = ScatterPlotItem.Symbols
symbols = numpy.array(list(Symbols.keys()))
shape_var = self.shapevar_model[shape_index]
data = column(self.data, shape_var).astype(numpy.float)
data = data % (len(Symbols) - 1)
data[numpy.isnan(data)] = len(Symbols) - 1
shape_data = symbols[data.astype(int)]
elif have_matrix_transposed and \
self.shapevar_model[shape_index] == 'Attribute names':
Symbols = ScatterPlotItem.Symbols
symbols = numpy.array(list(Symbols.keys()))
attr = [i % (len(Symbols) - 1)
for i, _ in enumerate(attributes(self.matrix))]
shape_data = symbols[attr]
else:
shape_data = "o"
self._shape_data = shape_data
if self._size_data is None:
MinPointSize = 3
point_size = self.symbol_size + MinPointSize
size_index = self.cb_size_value.currentIndex()
if have_data and size_index == 1:
# size by stress
size_data = stress(self.embedding, self._effective_matrix)
size_data = scale(size_data)
size_data = MinPointSize + size_data * point_size
elif have_data and size_index > 0:
size_var = self.sizevar_model[size_index]
size_data = column(self.data, size_var)
size_data = scale(size_data)
size_data = MinPointSize + size_data * point_size
else:
size_data = point_size
self._size_data = size_data
if self._label_data is None:
label_index = self.cb_label_value.currentIndex()
if have_data and label_index > 0:
label_var = self.labelvar_model[label_index]
label_data = column(self.data, label_var)
label_data = [label_var.str_val(val) for val in label_data]
label_items = [pg.TextItem(text, anchor=(0.5, 0), color=0.0)
for text in label_data]
elif have_matrix_transposed and \
self.labelvar_model[label_index] == 'Attribute names':
attr = attributes(self.matrix)
label_items = [pg.TextItem(str(text), anchor=(0.5, 0))
for text in attr]
else:
label_items = None
self._label_data = label_items
emb_x, emb_y = self.embedding[:, 0], self.embedding[:, 1]
if self.jitter > 0:
_, jitter_factor = self.JitterAmount[self.jitter]
emb_x = jitter(emb_x, jitter_factor, rstate=42)
emb_y = jitter(emb_y, jitter_factor, rstate=667)
if self.connected_pairs and self.__draw_similar_pairs:
if self._similar_pairs is None:
# This code requires storing lower triangle of X (n x n / 2
# doubles), n x n / 2 * 2 indices to X, n x n / 2 indices for
# argsort result. If this becomes an issue, it can be reduced to
# n x n argsort indices by argsorting the entire X. Then we
# take the first n + 2 * p indices. We compute their coordinates
# i, j in the original matrix. We keep those for which i < j.
# n + 2 * p will suffice to exclude the diagonal (i = j). If the
# number of those for which i < j is smaller than p, we instead
# take i > j. Among those that remain, we take the first p.
# Assuming that MDS can't show so many points that memory could
# become an issue, I preferred using simpler code.
m = self._effective_matrix
n = len(m)
p = (n * (n - 1) // 2 * self.connected_pairs) // 100
indcs = numpy.triu_indices(n, 1)
sorted = numpy.argsort(m[indcs])[:p]
self._similar_pairs = fpairs = numpy.empty(2 * p, dtype=int)
fpairs[::2] = indcs[0][sorted]
fpairs[1::2] = indcs[1][sorted]
for i in range(int(len(emb_x[self._similar_pairs]) / 2)):
item = QtGui.QGraphicsLineItem(
emb_x[self._similar_pairs][i * 2],
emb_y[self._similar_pairs][i * 2],
emb_x[self._similar_pairs][i * 2 + 1],
emb_y[self._similar_pairs][i * 2 + 1]
)
pen = QtGui.QPen(QtGui.QBrush(QtGui.QColor(204, 204, 204)), 2)
pen.setCosmetic(True)
item.setPen(pen)
self.plot.addItem(item)
data = numpy.arange(size)
self._scatter_item = item = ScatterPlotItem(
x=emb_x, y=emb_y,
pen=self._pen_data, brush=self._brush_data, symbol=self._shape_data,
size=self._size_data, data=data,
antialias=True
)
self.plot.addItem(item)
if self._label_data is not None:
if self.label_only_selected:
if self._selection_mask is not None:
for (x, y), text_item, selected \
in zip(self.embedding, self._label_data,
self._selection_mask):
if selected:
self.plot.addItem(text_item)
text_item.setPos(x, y)
else:
for (x, y), text_item in zip(self.embedding, self._label_data):
self.plot.addItem(text_item)
text_item.setPos(x, y)
self._legend_item = LegendItem()
viewbox = self.plot.getViewBox()
self._legend_item.setParentItem(self.plot.getViewBox())
self._legend_item.setZValue(viewbox.zValue() + 10)
self._legend_item.restoreAnchor(self.legend_anchor)
color_var = shape_var = None
color_index = self.cb_color_value.currentIndex()
if have_data and 1 <= color_index < len(self.colorvar_model):
color_var = self.colorvar_model[color_index]
assert isinstance(color_var, Orange.data.Variable)
shape_index = self.cb_shape_value.currentIndex()
if have_data and 1 <= shape_index < len(self.shapevar_model):
shape_var = self.shapevar_model[shape_index]
assert isinstance(shape_var, Orange.data.Variable)
if shape_var is not None or \
(color_var is not None and color_var.is_discrete):
legend_data = mdsplotutils.legend_data(
color_var, shape_var, plotstyle=plotstyle)
for color, symbol, text in legend_data:
self._legend_item.addItem(
ScatterPlotItem(pen=color, brush=color, symbol=symbol,
size=10),
escape(text)
)
else:
self._legend_item.hide()
def commit(self):
if self.embedding is not None:
output = embedding = Orange.data.Table.from_numpy(
Orange.data.Domain([Orange.data.ContinuousVariable("X"),
Orange.data.ContinuousVariable("Y")]),
self.embedding
)
else:
output = embedding = None
if self.embedding is not None and self.data is not None:
domain = self.data.domain
attrs = domain.attributes
class_vars = domain.class_vars
metas = domain.metas
if self.output_embedding_role == OWMDS.AttrRole:
attrs = embedding.domain.attributes
elif self.output_embedding_role == OWMDS.AddAttrRole:
attrs = domain.attributes + embedding.domain.attributes
elif self.output_embedding_role == OWMDS.MetaRole:
metas += embedding.domain.attributes
domain = Orange.data.Domain(attrs, class_vars, metas)
output = Orange.data.Table.from_table(domain, self.data)
if self.output_embedding_role == OWMDS.AttrRole:
output.X[:] = embedding.X
if self.output_embedding_role == OWMDS.AddAttrRole:
output.X[:, -2:] = embedding.X
elif self.output_embedding_role == OWMDS.MetaRole:
output.metas[:, -2:] = embedding.X
self.send("Data", output)
if output is not None and self._selection_mask is not None and \
numpy.any(self._selection_mask):
subset = output[self._selection_mask]
else:
subset = None
self.send("Selected Data", subset)
def onDeleteWidget(self):
super().onDeleteWidget()
self._clear_plot()
self._clear()
def __selection_end(self, path):
self.select(path)
self._pen_data = None
self._update_plot()
self._invalidate_output()
def select(self, region):
item = self._scatter_item
if item is None:
return
indices = numpy.array(
[spot.data() for spot in item.points()
if region.contains(spot.pos())],
dtype=int)
if not QtGui.QApplication.keyboardModifiers():
self._selection_mask = None
self.select_indices(indices, QtGui.QApplication.keyboardModifiers())
def select_indices(self, indices, modifiers=Qt.NoModifier):
if self.data is None:
return
if self._selection_mask is None or \
not modifiers & (Qt.ControlModifier | Qt.ShiftModifier |
Qt.AltModifier):
self._selection_mask = numpy.zeros(len(self.data), dtype=bool)
if modifiers & Qt.AltModifier:
self._selection_mask[indices] = False
elif modifiers & Qt.ControlModifier:
self._selection_mask[indices] = ~self._selection_mask[indices]
else:
self._selection_mask[indices] = True
def send_report(self):
if self.data is None:
return
self.report_plot()
caption = report.render_items_vert((
("Color", self.color_value != "Same color" and self.color_value),
("Shape", self.shape_value != "Same shape" and self.shape_value),
("Size", self.size_value != "Same size" and self.size_value),
("Labels", self.label_value != "No labels" and self.label_value)))
if caption:
self.report_caption(caption)
self.report_items((("Output", self.output_combo.currentText()),))
class OWDistributions(widget.OWWidget):
name = "Distributions"
description = "Display value distributions of a data feature in a graph."
icon = "icons/Distribution.svg"
priority = 100
inputs = [InputSignal("Data", Orange.data.Table, "set_data",
doc="Set the input data set")]
settingsHandler = settings.DomainContextHandler(
match_values=settings.DomainContextHandler.MATCH_VALUES_ALL)
#: Selected variable index
variable_idx = settings.ContextSetting(-1)
#: Selected group variable
groupvar_idx = settings.ContextSetting(0)
relative_freq = settings.Setting(False)
disc_cont = settings.Setting(False)
smoothing_index = settings.Setting(5)
show_prob = settings.ContextSetting(0)
want_graph = True
ASH_HIST = 50
bins = [ 2, 3, 4, 5, 8, 10, 12, 15, 20, 30, 50 ]
smoothing_facs = list(reversed([ 0.1, 0.2, 0.4, 0.6, 0.8, 1, 1.5, 2, 4, 6, 10 ]))
def __init__(self):
super().__init__()
self.data = None
self.distributions = None
self.contingencies = None
self.var = self.cvar = None
varbox = gui.widgetBox(self.controlArea, "Variable")
self.varmodel = itemmodels.VariableListModel()
self.groupvarmodel = []
self.varview = QtGui.QListView(
selectionMode=QtGui.QListView.SingleSelection)
self.varview.setSizePolicy(
QtGui.QSizePolicy.Minimum, QtGui.QSizePolicy.Expanding)
self.varview.setModel(self.varmodel)
self.varview.setSelectionModel(
itemmodels.ListSingleSelectionModel(self.varmodel))
self.varview.selectionModel().selectionChanged.connect(
self._on_variable_idx_changed)
varbox.layout().addWidget(self.varview)
box = gui.widgetBox(self.controlArea, "Precision")
gui.separator(self.controlArea, 4, 4)
box2 = gui.widgetBox(box, orientation="horizontal")
self.l_smoothing_l = gui.widgetLabel(box2, "Smooth")
gui.hSlider(box2, self, "smoothing_index",
minValue=0, maxValue=len(self.smoothing_facs) - 1,
callback=self._on_set_smoothing, createLabel=False)
self.l_smoothing_r = gui.widgetLabel(box2, "Precise")
self.cb_disc_cont = gui.checkBox(
gui.indentedBox(box, sep=4),
self, "disc_cont", "Bin continuous variables",
callback=self._on_groupvar_idx_changed)
box = gui.widgetBox(self.controlArea, "Group by")
self.icons = gui.attributeIconDict
self.groupvarview = gui.comboBox(box, self, "groupvar_idx",
callback=self._on_groupvar_idx_changed, valueType=str,
contentsLength=12)
box2 = gui.indentedBox(box, sep=4)
self.cb_rel_freq = gui.checkBox(
box2, self, "relative_freq", "Show relative frequencies",
callback=self._on_relative_freq_changed)
gui.separator(box2)
self.cb_prob = gui.comboBox(
box2, self, "show_prob", label="Show probabilities",
orientation="horizontal",
callback=self._on_relative_freq_changed)
plotview = pg.PlotWidget(background=None)
self.mainArea.layout().addWidget(plotview)
w = QtGui.QLabel()
w.setSizePolicy(QtGui.QSizePolicy.Expanding, QtGui.QSizePolicy.Fixed)
self.mainArea.layout().addWidget(w, Qt.AlignCenter)
self.plot = pg.PlotItem()
self.plot.hideButtons()
plotview.setCentralItem(self.plot)
self.plot_prob = pg.ViewBox()
self.plot.hideAxis('right')
self.plot.scene().addItem(self.plot_prob)
self.plot.getAxis("right").linkToView(self.plot_prob)
self.plot.getAxis("right").setLabel("Probability")
self.plot_prob.setZValue(10)
self.plot_prob.setXLink(self.plot)
self.update_views()
self.plot.vb.sigResized.connect(self.update_views)
self.plot_prob.setRange(yRange=[0,1])
def disable_mouse(plot):
plot.setMouseEnabled(False, False)
plot.setMenuEnabled(False)
disable_mouse(self.plot.getViewBox())
disable_mouse(self.plot_prob)
pen = QtGui.QPen(self.palette().color(QtGui.QPalette.Text))
for axis in ("left", "bottom"):
self.plot.getAxis(axis).setPen(pen)
self._legend = LegendItem()
self._legend.setParentItem(self.plot.getViewBox())
self._legend.hide()
self._legend.anchor((1, 0), (1, 0))
self.graphButton.clicked.connect(self.save_graph)
def update_views(self):
self.plot_prob.setGeometry(self.plot.vb.sceneBoundingRect())
self.plot_prob.linkedViewChanged(self.plot.vb, self.plot_prob.XAxis)
def set_data(self, data):
self.closeContext()
self.clear()
self.data = data
if self.data is not None:
domain = self.data.domain
self.varmodel[:] = list(domain)
self.groupvarview.clear()
self.groupvarmodel = \
["(None)"] + [var for var in domain if var.is_discrete]
self.groupvarview.addItem("(None)")
for var in self.groupvarmodel[1:]:
self.groupvarview.addItem(self.icons[var], var.name)
if domain.has_discrete_class:
self.groupvar_idx = \
self.groupvarmodel.index(domain.class_var)
self.openContext(domain)
self.variable_idx = min(max(self.variable_idx, 0),
len(self.varmodel) - 1)
self.groupvar_idx = min(max(self.groupvar_idx, 0),
len(self.groupvarmodel) - 1)
itemmodels.select_row(self.varview, self.variable_idx)
self._setup()
def clear(self):
self.plot.clear()
self.plot_prob.clear()
self.varmodel[:] = []
self.groupvarmodel = []
self.variable_idx = -1
self.groupvar_idx = 0
self._legend.clear()
self._legend.hide()
def _setup_smoothing(self):
if not self.disc_cont and self.var and self.var.is_continuous:
self.cb_disc_cont.setText("Bin continuous variables")
self.l_smoothing_l.setText("Smooth")
self.l_smoothing_r.setText("Precise")
else:
self.cb_disc_cont.setText("Bin continuous variables into {} bins".
format(self.bins[self.smoothing_index]))
self.l_smoothing_l.setText(" " + str(self.bins[0]))
self.l_smoothing_r.setText(" " + str(self.bins[-1]))
def _setup(self):
self.plot.clear()
self.plot_prob.clear()
self._legend.clear()
self._legend.hide()
varidx = self.variable_idx
self.var = self.cvar = None
if varidx >= 0:
self.var = self.varmodel[varidx]
if self.groupvar_idx > 0:
self.cvar = self.groupvarmodel[self.groupvar_idx]
self.cb_prob.clear()
self.cb_prob.addItem("(None)")
self.cb_prob.addItems(self.cvar.values)
self.cb_prob.addItem("(All)")
self.show_prob = min(max(self.show_prob, 0),
len(self.cvar.values) + 1)
data = self.data
self._setup_smoothing()
if self.var is None:
return
if self.disc_cont:
data = self.data[:, (self.var, self.cvar) if self.cvar else self.var ]
disc = Orange.preprocess.discretize.EqualWidth(n=self.bins[self.smoothing_index])
data = Orange.preprocess.Discretize(data, method=disc)
self.var = data.domain.variables[0]
self.set_left_axis_name()
self.enable_disable_rel_freq()
if self.cvar:
self.contingencies = \
contingency.get_contingency(data, self.var, self.cvar)
self.display_contingency()
else:
self.distributions = \
distribution.get_distribution(data, self.var)
self.display_distribution()
self.plot.autoRange()
def display_distribution(self):
dist = self.distributions
var = self.var
assert len(dist) > 0
self.plot.clear()
self.plot_prob.clear()
self.plot.hideAxis('right')
bottomaxis = self.plot.getAxis("bottom")
bottomaxis.setLabel(var.name)
self.set_left_axis_name()
if var and var.is_continuous:
bottomaxis.setTicks(None)
edges, curve = ash_curve(dist, None, m=OWDistributions.ASH_HIST,
smoothing_factor=self.smoothing_facs[self.smoothing_index])
edges = edges + (edges[1] - edges[0])/2
edges = edges[:-1]
item = pg.PlotCurveItem()
pen = QtGui.QPen(QtGui.QBrush(Qt.white), 3)
pen.setCosmetic(True)
item.setData(edges, curve, antialias=True, stepMode=False,
fillLevel=0, brush=QtGui.QBrush(Qt.gray),
pen=pen)
self.plot.addItem(item)
else:
bottomaxis.setTicks([list(enumerate(var.values))])
for i, w in enumerate(dist):
geom = QtCore.QRectF(i - 0.33, 0, 0.66, w)
item = DistributionBarItem(geom, [1.0],
[QtGui.QColor(128, 128, 128)])
self.plot.addItem(item)
def _on_relative_freq_changed(self):
self.set_left_axis_name()
if self.cvar and self.cvar.is_discrete:
self.display_contingency()
else:
self.display_distribution()
self.plot.autoRange()
def display_contingency(self):
"""
Set the contingency to display.
"""
cont = self.contingencies
var, cvar = self.var, self.cvar
assert len(cont) > 0
self.plot.clear()
self.plot_prob.clear()
self._legend.clear()
if self.show_prob:
self.plot.showAxis('right')
else:
self.plot.hideAxis('right')
bottomaxis = self.plot.getAxis("bottom")
bottomaxis.setLabel(var.name)
cvar_values = cvar.values
palette = colorpalette.ColorPaletteGenerator(len(cvar_values))
colors = [palette[i].lighter() for i in range(len(cvar_values))]
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_facs[self.smoothing_index]))
weights = numpy.array(weights)
sumw = numpy.sum(weights)
weights /= sumw
colors = cols
curves = [(X, Y * w) for (X, Y), w in zip(curves, weights)]
ncval = len(cvar_values)
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" ]:
for (X, Y), color, w in reversed(list(zip(curvesline, colors, weights))):
item = pg.PlotCurveItem()
pen = QtGui.QPen(QtGui.QBrush(color), 3)
pen.setCosmetic(True)
color = QtGui.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=QtGui.QBrush(color), pen=pen)
self.plot.addItem(item)
if self.show_prob:
M_EST = 5 #for M estimate
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)
# allcorrection = M_EST/sumw*numpy.sum(sumprob)/len(inter_X)
legal = sumprob > 0.05 * numpy.max(sumprob)
i = len(curvesinterp) + 1
show_all = self.show_prob == i
for Y, color in reversed(list(zip(curvesinterp, colors))):
i -= 1
if show_all or self.show_prob == i:
item = pg.PlotCurveItem()
pen = QtGui.QPen(QtGui.QBrush(color), 3, style=QtCore.Qt.DotLine)
pen.setCosmetic(True)
#prob = (Y+allcorrection/ncval)/(sumprob+allcorrection)
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)
elif var and var.is_discrete:
bottomaxis.setTicks([list(enumerate(var.values))])
cont = numpy.array(cont)
ncval = len(cvar_values)
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 = QtCore.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)
if self.show_prob:
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)
mark = pg.ScatterPlotItem()
bar = pg.ErrorBarItem()
pen = QtGui.QPen(QtGui.QBrush(QtGui.QColor(0)), 1)
pen.setCosmetic(True)
bar.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=QtGui.QColor(1), pen=pen)
mark.setData([i+position], [prob], antialias=True, symbol="o",
fillLevel=None, pxMode=True, size=10,
brush=QtGui.QColor(colors[ic]), pen=pen)
self.plot_prob.addItem(bar)
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()
def set_left_axis_name(self):
set_label = self.plot.getAxis("left").setLabel
if self.var and self.var.is_continuous:
set_label(["Density", "Relative density"]
[self.cvar is not None and self.relative_freq])
else:
set_label(["Frequency", "Relative frequency"]
[self.cvar is not None and self.relative_freq])
def enable_disable_rel_freq(self):
self.cb_prob.setDisabled(self.var is None or self.cvar is None)
self.cb_rel_freq.setDisabled(
self.var is None or self.cvar is None)
def _on_variable_idx_changed(self):
self.variable_idx = selected_index(self.varview)
self._setup()
def _on_groupvar_idx_changed(self):
self._setup()
def _on_set_smoothing(self):
self._setup()
def onDeleteWidget(self):
self.plot.clear()
super().onDeleteWidget()
def save_graph(self):
from Orange.widgets.data.owsave import OWSave
save_img = OWSave(data=self.plot,
file_formats=FileFormat.img_writers)
save_img.exec_()
def _setup_plot(self):
have_data = self.data is not None
have_matrix_transposed = self.matrix is not None and not self.matrix.axis
plotstyle = mdsplotutils.plotstyle
size = self._effective_matrix.shape[0]
def column(data, variable):
a, _ = data.get_column_view(variable)
return a.ravel()
def attributes(matrix):
return matrix.row_items.domain.attributes
def scale(a):
dmin, dmax = numpy.nanmin(a), numpy.nanmax(a)
if dmax - dmin > 0:
return (a - dmin) / (dmax - dmin)
else:
return numpy.zeros_like(a)
def jitter(x, factor=1, rstate=None):
if rstate is None:
rstate = numpy.random.RandomState()
elif not isinstance(rstate, numpy.random.RandomState):
rstate = numpy.random.RandomState(rstate)
span = numpy.nanmax(x) - numpy.nanmin(x)
if span < numpy.finfo(x.dtype).eps * 100:
span = 1
a = factor * span / 100.
return x + (rstate.random_sample(x.shape) - 0.5) * a
if self._pen_data is None:
if self._selection_mask is not None:
pointflags = numpy.where(
self._selection_mask,
mdsplotutils.Selected, mdsplotutils.NoFlags)
else:
pointflags = None
color_index = self.cb_color_value.currentIndex()
if have_data and color_index > 0:
color_var = self.colorvar_model[color_index]
if color_var.is_discrete:
palette = colorpalette.ColorPaletteGenerator(
len(color_var.values)
)
plotstyle = plotstyle.updated(discrete_palette=palette)
else:
palette = None
color_data = mdsplotutils.color_data(
self.data, color_var, plotstyle=plotstyle)
color_data = numpy.hstack(
(color_data,
numpy.full((len(color_data), 1), self.symbol_opacity,
dtype=float))
)
pen_data = mdsplotutils.pen_data(color_data * 0.8, pointflags)
brush_data = mdsplotutils.brush_data(color_data)
elif have_matrix_transposed and \
self.colorvar_model[color_index] == 'Attribute names':
attr = attributes(self.matrix)
palette = colorpalette.ColorPaletteGenerator(len(attr))
color_data = [palette.getRGB(i) for i in range(len(attr))]
color_data = numpy.hstack((
color_data,
numpy.full((len(color_data), 1), self.symbol_opacity,
dtype=float))
)
pen_data = mdsplotutils.pen_data(color_data * 0.8, pointflags)
brush_data = mdsplotutils.brush_data(color_data)
else:
pen_data = make_pen(QtGui.QColor(Qt.darkGray), cosmetic=True)
if self._selection_mask is not None:
pen_data = numpy.array(
[pen_data, plotstyle.selected_pen])
pen_data = pen_data[self._selection_mask.astype(int)]
else:
pen_data = numpy.full(self._effective_matrix.dim, pen_data,
dtype=object)
brush_data = numpy.full(
size, pg.mkColor((192, 192, 192, self.symbol_opacity)),
dtype=object)
if self._subset_mask is not None and have_data and \
self._subset_mask.shape == (size, ):
# clear brush fill for non subset data
brush_data[~self._subset_mask] = QtGui.QBrush(Qt.NoBrush)
self._pen_data = pen_data
self._brush_data = brush_data
if self._shape_data is None:
shape_index = self.cb_shape_value.currentIndex()
if have_data and shape_index > 0:
Symbols = ScatterPlotItem.Symbols
symbols = numpy.array(list(Symbols.keys()))
shape_var = self.shapevar_model[shape_index]
data = column(self.data, shape_var).astype(numpy.float)
data = data % (len(Symbols) - 1)
data[numpy.isnan(data)] = len(Symbols) - 1
shape_data = symbols[data.astype(int)]
elif have_matrix_transposed and \
self.shapevar_model[shape_index] == 'Attribute names':
Symbols = ScatterPlotItem.Symbols
symbols = numpy.array(list(Symbols.keys()))
attr = [i % (len(Symbols) - 1)
for i, _ in enumerate(attributes(self.matrix))]
shape_data = symbols[attr]
else:
shape_data = "o"
self._shape_data = shape_data
if self._size_data is None:
MinPointSize = 3
point_size = self.symbol_size + MinPointSize
size_index = self.cb_size_value.currentIndex()
if have_data and size_index == 1:
# size by stress
size_data = stress(self.embedding, self._effective_matrix)
size_data = scale(size_data)
size_data = MinPointSize + size_data * point_size
elif have_data and size_index > 0:
size_var = self.sizevar_model[size_index]
size_data = column(self.data, size_var)
size_data = scale(size_data)
size_data = MinPointSize + size_data * point_size
else:
size_data = point_size
self._size_data = size_data
if self._label_data is None:
label_index = self.cb_label_value.currentIndex()
if have_data and label_index > 0:
label_var = self.labelvar_model[label_index]
label_data = column(self.data, label_var)
label_data = [label_var.str_val(val) for val in label_data]
label_items = [pg.TextItem(text, anchor=(0.5, 0), color=0.0)
for text in label_data]
elif have_matrix_transposed and \
self.labelvar_model[label_index] == 'Attribute names':
attr = attributes(self.matrix)
label_items = [pg.TextItem(str(text), anchor=(0.5, 0))
for text in attr]
else:
label_items = None
self._label_data = label_items
emb_x, emb_y = self.embedding[:, 0], self.embedding[:, 1]
if self.jitter > 0:
_, jitter_factor = self.JitterAmount[self.jitter]
emb_x = jitter(emb_x, jitter_factor, rstate=42)
emb_y = jitter(emb_y, jitter_factor, rstate=667)
if self.connected_pairs and self.__draw_similar_pairs:
if self._similar_pairs is None:
# This code requires storing lower triangle of X (n x n / 2
# doubles), n x n / 2 * 2 indices to X, n x n / 2 indices for
# argsort result. If this becomes an issue, it can be reduced to
# n x n argsort indices by argsorting the entire X. Then we
# take the first n + 2 * p indices. We compute their coordinates
# i, j in the original matrix. We keep those for which i < j.
# n + 2 * p will suffice to exclude the diagonal (i = j). If the
# number of those for which i < j is smaller than p, we instead
# take i > j. Among those that remain, we take the first p.
# Assuming that MDS can't show so many points that memory could
# become an issue, I preferred using simpler code.
m = self._effective_matrix
n = len(m)
p = (n * (n - 1) // 2 * self.connected_pairs) // 100
indcs = numpy.triu_indices(n, 1)
sorted = numpy.argsort(m[indcs])[:p]
self._similar_pairs = fpairs = numpy.empty(2 * p, dtype=int)
fpairs[::2] = indcs[0][sorted]
fpairs[1::2] = indcs[1][sorted]
for i in range(int(len(emb_x[self._similar_pairs]) / 2)):
item = QtGui.QGraphicsLineItem(
emb_x[self._similar_pairs][i * 2],
emb_y[self._similar_pairs][i * 2],
emb_x[self._similar_pairs][i * 2 + 1],
emb_y[self._similar_pairs][i * 2 + 1]
)
pen = QtGui.QPen(QtGui.QBrush(QtGui.QColor(204, 204, 204)), 2)
pen.setCosmetic(True)
item.setPen(pen)
self.plot.addItem(item)
data = numpy.arange(size)
self._scatter_item = item = ScatterPlotItem(
x=emb_x, y=emb_y,
pen=self._pen_data, brush=self._brush_data, symbol=self._shape_data,
size=self._size_data, data=data,
antialias=True
)
self.plot.addItem(item)
if self._label_data is not None:
if self.label_only_selected:
if self._selection_mask is not None:
for (x, y), text_item, selected \
in zip(self.embedding, self._label_data,
self._selection_mask):
if selected:
self.plot.addItem(text_item)
text_item.setPos(x, y)
else:
for (x, y), text_item in zip(self.embedding, self._label_data):
self.plot.addItem(text_item)
text_item.setPos(x, y)
self._legend_item = LegendItem()
viewbox = self.plot.getViewBox()
self._legend_item.setParentItem(self.plot.getViewBox())
self._legend_item.setZValue(viewbox.zValue() + 10)
self._legend_item.restoreAnchor(self.legend_anchor)
color_var = shape_var = None
color_index = self.cb_color_value.currentIndex()
if have_data and 1 <= color_index < len(self.colorvar_model):
color_var = self.colorvar_model[color_index]
assert isinstance(color_var, Orange.data.Variable)
shape_index = self.cb_shape_value.currentIndex()
if have_data and 1 <= shape_index < len(self.shapevar_model):
shape_var = self.shapevar_model[shape_index]
assert isinstance(shape_var, Orange.data.Variable)
if shape_var is not None or \
(color_var is not None and color_var.is_discrete):
legend_data = mdsplotutils.legend_data(
color_var, shape_var, plotstyle=plotstyle)
for color, symbol, text in legend_data:
self._legend_item.addItem(
ScatterPlotItem(pen=color, brush=color, symbol=symbol,
size=10),
escape(text)
)
else:
self._legend_item.hide()
class OWDistributions(widget.OWWidget):
name = "Distributions"
description = "Display value distributions of a data feature in a graph."
icon = "icons/Distribution.svg"
priority = 120
class Inputs:
data = Input("Data", Orange.data.Table, doc="Set the input data set")
settingsHandler = settings.DomainContextHandler(
match_values=settings.DomainContextHandler.MATCH_VALUES_ALL)
#: Selected variable index
variable_idx = settings.ContextSetting(-1)
#: Selected group variable
groupvar_idx = settings.ContextSetting(0)
relative_freq = settings.Setting(False)
disc_cont = settings.Setting(False)
smoothing_index = settings.Setting(5)
show_prob = settings.ContextSetting(0)
graph_name = "plot"
ASH_HIST = 50
bins = [2, 3, 4, 5, 8, 10, 12, 15, 20, 30, 50]
smoothing_facs = list(reversed([0.1, 0.2, 0.4, 0.6, 0.8, 1, 1.5, 2, 4, 6, 10]))
def __init__(self):
super().__init__()
self.data = None
self.distributions = None
self.contingencies = None
self.var = self.cvar = None
varbox = gui.vBox(self.controlArea, "Variable")
self.varmodel = itemmodels.VariableListModel()
self.groupvarmodel = []
self.varview = QListView(
selectionMode=QListView.SingleSelection)
self.varview.setSizePolicy(
QSizePolicy.Minimum, QSizePolicy.Expanding)
self.varview.setModel(self.varmodel)
self.varview.setSelectionModel(
itemmodels.ListSingleSelectionModel(self.varmodel))
self.varview.selectionModel().selectionChanged.connect(
self._on_variable_idx_changed)
varbox.layout().addWidget(self.varview)
box = gui.vBox(self.controlArea, "Precision")
gui.separator(self.controlArea, 4, 4)
box2 = gui.hBox(box)
self.l_smoothing_l = gui.widgetLabel(box2, "Smooth")
gui.hSlider(box2, self, "smoothing_index",
minValue=0, maxValue=len(self.smoothing_facs) - 1,
callback=self._on_set_smoothing, createLabel=False)
self.l_smoothing_r = gui.widgetLabel(box2, "Precise")
self.cb_disc_cont = gui.checkBox(
gui.indentedBox(box, sep=4),
self, "disc_cont", "Bin numeric variables",
callback=self._on_groupvar_idx_changed,
tooltip="Show numeric variables as categorical.")
box = gui.vBox(self.controlArea, "Group by")
self.icons = gui.attributeIconDict
self.groupvarview = gui.comboBox(
box, self, "groupvar_idx",
callback=self._on_groupvar_idx_changed,
valueType=str, contentsLength=12)
box2 = gui.indentedBox(box, sep=4)
self.cb_rel_freq = gui.checkBox(
box2, self, "relative_freq", "Show relative frequencies",
callback=self._on_relative_freq_changed,
tooltip="Normalize probabilities so that probabilities "
"for each group-by value sum to 1.")
gui.separator(box2)
self.cb_prob = gui.comboBox(
box2, self, "show_prob", label="Show probabilities:",
orientation=Qt.Horizontal,
callback=self._on_relative_freq_changed,
tooltip="Show probabilities for a chosen group-by value "
"(at each point probabilities for all group-by values sum to 1).")
self.plotview = pg.PlotWidget(background=None)
self.plotview.setRenderHint(QPainter.Antialiasing)
self.mainArea.layout().addWidget(self.plotview)
w = QLabel()
w.setSizePolicy(QSizePolicy.Expanding, QSizePolicy.Fixed)
self.mainArea.layout().addWidget(w, Qt.AlignCenter)
self.ploti = pg.PlotItem()
self.plot = self.ploti.vb
self.ploti.hideButtons()
self.plotview.setCentralItem(self.ploti)
self.plot_prob = pg.ViewBox()
self.ploti.hideAxis('right')
self.ploti.scene().addItem(self.plot_prob)
self.ploti.getAxis("right").linkToView(self.plot_prob)
self.ploti.getAxis("right").setLabel("Probability")
self.plot_prob.setZValue(10)
self.plot_prob.setXLink(self.ploti)
self.update_views()
self.ploti.vb.sigResized.connect(self.update_views)
self.plot_prob.setRange(yRange=[0, 1])
def disable_mouse(plot):
plot.setMouseEnabled(False, False)
plot.setMenuEnabled(False)
disable_mouse(self.plot)
disable_mouse(self.plot_prob)
self.tooltip_items = []
self.plot.scene().installEventFilter(
HelpEventDelegate(self.help_event, self))
pen = QPen(self.palette().color(QPalette.Text))
for axis in ("left", "bottom"):
self.ploti.getAxis(axis).setPen(pen)
self._legend = LegendItem()
self._legend.setParentItem(self.plot)
self._legend.hide()
self._legend.anchor((1, 0), (1, 0))
def update_views(self):
self.plot_prob.setGeometry(self.plot.sceneBoundingRect())
self.plot_prob.linkedViewChanged(self.plot, self.plot_prob.XAxis)
@Inputs.data
def set_data(self, data):
self.closeContext()
self.clear()
self.warning()
self.data = data
self.distributions = None
self.contingencies = None
if self.data is not None:
if not self.data:
self.warning("Empty input data cannot be visualized")
return
domain = self.data.domain
self.varmodel[:] = list(domain.variables) + \
[meta for meta in domain.metas
if meta.is_continuous or meta.is_discrete]
self.groupvarview.clear()
self.groupvarmodel = \
["(None)"] + [var for var in domain.variables if var.is_discrete] + \
[meta for meta in domain.metas if meta.is_discrete]
self.groupvarview.addItem("(None)")
for var in self.groupvarmodel[1:]:
self.groupvarview.addItem(self.icons[var], var.name)
if domain.has_discrete_class:
self.groupvar_idx = \
self.groupvarmodel[1:].index(domain.class_var) + 1
self.openContext(domain)
self.variable_idx = min(max(self.variable_idx, 0),
len(self.varmodel) - 1)
self.groupvar_idx = min(max(self.groupvar_idx, 0),
len(self.groupvarmodel) - 1)
itemmodels.select_row(self.varview, self.variable_idx)
self._setup()
def clear(self):
self.plot.clear()
self.plot_prob.clear()
self.varmodel[:] = []
self.groupvarmodel = []
self.variable_idx = -1
self.groupvar_idx = 0
self._legend.clear()
self._legend.hide()
self.groupvarview.clear()
self.cb_prob.clear()
def _setup_smoothing(self):
if not self.disc_cont and self.var and self.var.is_continuous:
self.cb_disc_cont.setText("Bin numeric variables")
self.l_smoothing_l.setText("Smooth")
self.l_smoothing_r.setText("Precise")
else:
self.cb_disc_cont.setText("Bin numeric variables into {} bins".
format(self.bins[self.smoothing_index]))
self.l_smoothing_l.setText(" " + str(self.bins[0]))
self.l_smoothing_r.setText(" " + str(self.bins[-1]))
@property
def smoothing_factor(self):
return self.smoothing_facs[self.smoothing_index]
def _setup(self):
self.plot.clear()
self.plot_prob.clear()
self._legend.clear()
self._legend.hide()
varidx = self.variable_idx
self.var = self.cvar = None
if varidx >= 0:
self.var = self.varmodel[varidx]
if self.groupvar_idx > 0:
self.cvar = self.groupvarmodel[self.groupvar_idx]
self.cb_prob.clear()
self.cb_prob.addItem("(None)")
self.cb_prob.addItems(self.cvar.values)
self.cb_prob.addItem("(All)")
self.show_prob = min(max(self.show_prob, 0),
len(self.cvar.values) + 1)
data = self.data
self._setup_smoothing()
if self.var is None:
return
if self.disc_cont:
domain = Orange.data.Domain(
[self.var, self.cvar] if self.cvar else [self.var])
data = Orange.data.Table(domain, data)
disc = EqualWidth(n=self.bins[self.smoothing_index])
data = Discretize(method=disc, remove_const=False)(data)
self.var = data.domain[0]
self.set_left_axis_name()
self.enable_disable_rel_freq()
if self.cvar:
self.contingencies = \
contingency.get_contingency(data, self.var, self.cvar)
self.display_contingency()
else:
self.distributions = \
distribution.get_distribution(data, self.var)
self.display_distribution()
self.plot.autoRange()
def help_event(self, ev):
self.plot.mapSceneToView(ev.scenePos())
ctooltip = []
for vb, item in self.tooltip_items:
mouse_over_curve = isinstance(item, pg.PlotCurveItem) \
and item.mouseShape().contains(vb.mapSceneToView(ev.scenePos()))
mouse_over_bar = isinstance(item, DistributionBarItem) \
and item.boundingRect().contains(vb.mapSceneToView(ev.scenePos()))
if mouse_over_curve or mouse_over_bar:
ctooltip.append(item.tooltip)
if ctooltip:
QToolTip.showText(ev.screenPos(), "\n\n".join(ctooltip), widget=self.plotview)
return True
return False
def display_distribution(self):
dist = self.distributions
var = self.var
if dist is None or not len(dist):
return
self.plot.clear()
self.plot_prob.clear()
self.ploti.hideAxis('right')
self.tooltip_items = []
bottomaxis = self.ploti.getAxis("bottom")
bottomaxis.setLabel(var.name)
bottomaxis.resizeEvent()
self.set_left_axis_name()
if var and var.is_continuous:
bottomaxis.setTicks(None)
if not len(dist[0]):
return
edges, curve = ash_curve(dist, None, m=OWDistributions.ASH_HIST,
smoothing_factor=self.smoothing_factor)
edges = edges + (edges[1] - edges[0])/2
edges = edges[:-1]
item = pg.PlotCurveItem()
pen = QPen(QBrush(Qt.white), 3)
pen.setCosmetic(True)
item.setData(edges, curve, antialias=True, stepMode=False,
fillLevel=0, brush=QBrush(Qt.gray), pen=pen)
self.plot.addItem(item)
item.tooltip = "Density"
self.tooltip_items.append((self.plot, item))
else:
bottomaxis.setTicks([list(enumerate(var.values))])
for i, w in enumerate(dist):
geom = QRectF(i - 0.33, 0, 0.66, w)
item = DistributionBarItem(geom, [1.0],
[QColor(128, 128, 128)])
self.plot.addItem(item)
item.tooltip = "Frequency for %s: %r" % (var.values[i], w)
self.tooltip_items.append((self.plot, item))
def _on_relative_freq_changed(self):
self.set_left_axis_name()
if self.cvar and self.cvar.is_discrete:
self.display_contingency()
else:
self.display_distribution()
self.plot.autoRange()
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()
def set_left_axis_name(self):
leftaxis = self.ploti.getAxis("left")
set_label = leftaxis.setLabel
if self.var and self.var.is_continuous:
set_label(["Density", "Relative density"]
[self.cvar is not None and self.relative_freq])
else:
set_label(["Frequency", "Relative frequency"]
[self.cvar is not None and self.relative_freq])
leftaxis.resizeEvent()
def enable_disable_rel_freq(self):
self.cb_prob.setDisabled(self.var is None or self.cvar is None)
self.cb_rel_freq.setDisabled(
self.var is None or self.cvar is None)
def _on_variable_idx_changed(self):
self.variable_idx = selected_index(self.varview)
self._setup()
def _on_groupvar_idx_changed(self):
self._setup()
def _on_set_smoothing(self):
self._setup()
def onDeleteWidget(self):
self.plot.clear()
super().onDeleteWidget()
def get_widget_name_extension(self):
if self.variable_idx >= 0:
return self.varmodel[self.variable_idx]
def send_report(self):
self.plotview.scene().setSceneRect(self.plotview.sceneRect())
if self.variable_idx < 0:
return
self.report_plot()
text = "Distribution of '{}'".format(
self.varmodel[self.variable_idx])
if self.groupvar_idx:
group_var = self.groupvarmodel[self.groupvar_idx]
prob = self.cb_prob
indiv_probs = 0 < prob.currentIndex() < prob.count() - 1
if not indiv_probs or self.relative_freq:
text += " grouped by '{}'".format(group_var)
if self.relative_freq:
text += " (relative frequencies)"
if indiv_probs:
text += "; probabilites for '{}={}'".format(
group_var, prob.currentText())
self.report_caption(text)
def __init__(self):
super().__init__()
self.data = None
self.distributions = None
self.contingencies = None
self.var = self.cvar = None
varbox = gui.widgetBox(self.controlArea, "Variable")
self.varmodel = itemmodels.VariableListModel()
self.groupvarmodel = []
self.varview = QtGui.QListView(
selectionMode=QtGui.QListView.SingleSelection)
self.varview.setSizePolicy(
QtGui.QSizePolicy.Minimum, QtGui.QSizePolicy.Expanding)
self.varview.setModel(self.varmodel)
self.varview.setSelectionModel(
itemmodels.ListSingleSelectionModel(self.varmodel))
self.varview.selectionModel().selectionChanged.connect(
self._on_variable_idx_changed)
varbox.layout().addWidget(self.varview)
box = gui.widgetBox(self.controlArea, "Precision")
gui.separator(self.controlArea, 4, 4)
box2 = gui.widgetBox(box, orientation="horizontal")
self.l_smoothing_l = gui.widgetLabel(box2, "Smooth")
gui.hSlider(box2, self, "smoothing_index",
minValue=0, maxValue=len(self.smoothing_facs) - 1,
callback=self._on_set_smoothing, createLabel=False)
self.l_smoothing_r = gui.widgetLabel(box2, "Precise")
self.cb_disc_cont = gui.checkBox(
gui.indentedBox(box, sep=4),
self, "disc_cont", "Bin continuous variables",
callback=self._on_groupvar_idx_changed)
box = gui.widgetBox(self.controlArea, "Group by")
self.icons = gui.attributeIconDict
self.groupvarview = gui.comboBox(box, self, "groupvar_idx",
callback=self._on_groupvar_idx_changed, valueType=str,
contentsLength=12)
box2 = gui.indentedBox(box, sep=4)
self.cb_rel_freq = gui.checkBox(
box2, self, "relative_freq", "Show relative frequencies",
callback=self._on_relative_freq_changed)
gui.separator(box2)
self.cb_prob = gui.comboBox(
box2, self, "show_prob", label="Show probabilities",
orientation="horizontal",
callback=self._on_relative_freq_changed)
plotview = pg.PlotWidget(background=None)
self.mainArea.layout().addWidget(plotview)
w = QtGui.QLabel()
w.setSizePolicy(QtGui.QSizePolicy.Expanding, QtGui.QSizePolicy.Fixed)
self.mainArea.layout().addWidget(w, Qt.AlignCenter)
self.plot = pg.PlotItem()
self.plot.hideButtons()
plotview.setCentralItem(self.plot)
self.plot_prob = pg.ViewBox()
self.plot.hideAxis('right')
self.plot.scene().addItem(self.plot_prob)
self.plot.getAxis("right").linkToView(self.plot_prob)
self.plot.getAxis("right").setLabel("Probability")
self.plot_prob.setZValue(10)
self.plot_prob.setXLink(self.plot)
self.update_views()
self.plot.vb.sigResized.connect(self.update_views)
self.plot_prob.setRange(yRange=[0,1])
def disable_mouse(plot):
plot.setMouseEnabled(False, False)
plot.setMenuEnabled(False)
disable_mouse(self.plot.getViewBox())
disable_mouse(self.plot_prob)
pen = QtGui.QPen(self.palette().color(QtGui.QPalette.Text))
for axis in ("left", "bottom"):
self.plot.getAxis(axis).setPen(pen)
self._legend = LegendItem()
self._legend.setParentItem(self.plot.getViewBox())
self._legend.hide()
self._legend.anchor((1, 0), (1, 0))
self.graphButton.clicked.connect(self.save_graph)
class OWDistributions(widget.OWWidget):
name = "Distributions"
description = "Display value distributions of a data feature in a graph."
icon = "icons/Distribution.svg"
priority = 100
inputs = [
InputSignal("Data",
Orange.data.Table,
"set_data",
doc="Set the input data set")
]
settingsHandler = settings.DomainContextHandler(
match_values=settings.DomainContextHandler.MATCH_VALUES_ALL)
#: Selected variable index
variable_idx = settings.ContextSetting(-1)
#: Selected group variable
groupvar_idx = settings.ContextSetting(0)
relative_freq = settings.Setting(False)
disc_cont = settings.Setting(False)
smoothing_index = settings.Setting(5)
show_prob = settings.ContextSetting(0)
graph_name = "plot"
ASH_HIST = 50
bins = [2, 3, 4, 5, 8, 10, 12, 15, 20, 30, 50]
smoothing_facs = list(
reversed([0.1, 0.2, 0.4, 0.6, 0.8, 1, 1.5, 2, 4, 6, 10]))
def __init__(self):
super().__init__()
self.data = None
self.distributions = None
self.contingencies = None
self.var = self.cvar = None
varbox = gui.widgetBox(self.controlArea, "Variable")
self.varmodel = itemmodels.VariableListModel()
self.groupvarmodel = []
self.varview = QtGui.QListView(
selectionMode=QtGui.QListView.SingleSelection)
self.varview.setSizePolicy(QtGui.QSizePolicy.Minimum,
QtGui.QSizePolicy.Expanding)
self.varview.setModel(self.varmodel)
self.varview.setSelectionModel(
itemmodels.ListSingleSelectionModel(self.varmodel))
self.varview.selectionModel().selectionChanged.connect(
self._on_variable_idx_changed)
varbox.layout().addWidget(self.varview)
box = gui.widgetBox(self.controlArea, "Precision")
gui.separator(self.controlArea, 4, 4)
box2 = gui.widgetBox(box, orientation="horizontal")
self.l_smoothing_l = gui.widgetLabel(box2, "Smooth")
gui.hSlider(box2,
self,
"smoothing_index",
minValue=0,
maxValue=len(self.smoothing_facs) - 1,
callback=self._on_set_smoothing,
createLabel=False)
self.l_smoothing_r = gui.widgetLabel(box2, "Precise")
self.cb_disc_cont = gui.checkBox(
gui.indentedBox(box, sep=4),
self,
"disc_cont",
"Bin continuous variables",
callback=self._on_groupvar_idx_changed,
tooltip="Show continuous variables as discrete.")
box = gui.widgetBox(self.controlArea, "Group by")
self.icons = gui.attributeIconDict
self.groupvarview = gui.comboBox(
box,
self,
"groupvar_idx",
callback=self._on_groupvar_idx_changed,
valueType=str,
contentsLength=12)
box2 = gui.indentedBox(box, sep=4)
self.cb_rel_freq = gui.checkBox(
box2,
self,
"relative_freq",
"Show relative frequencies",
callback=self._on_relative_freq_changed,
tooltip=
"Normalize probabilities so that probabilities for each group-by value sum to 1."
)
gui.separator(box2)
self.cb_prob = gui.comboBox(
box2,
self,
"show_prob",
label="Show probabilities",
orientation="horizontal",
callback=self._on_relative_freq_changed,
tooltip=
"Show probabilities for a chosen group-by value (at each point probabilities for all group-by values sum to 1)."
)
self.plotview = pg.PlotWidget(background=None)
self.plotview.setRenderHint(QtGui.QPainter.Antialiasing)
self.mainArea.layout().addWidget(self.plotview)
w = QtGui.QLabel()
w.setSizePolicy(QtGui.QSizePolicy.Expanding, QtGui.QSizePolicy.Fixed)
self.mainArea.layout().addWidget(w, Qt.AlignCenter)
self.ploti = pg.PlotItem()
self.plot = self.ploti.vb
self.ploti.hideButtons()
self.plotview.setCentralItem(self.ploti)
self.plot_prob = pg.ViewBox()
self.ploti.hideAxis('right')
self.ploti.scene().addItem(self.plot_prob)
self.ploti.getAxis("right").linkToView(self.plot_prob)
self.ploti.getAxis("right").setLabel("Probability")
self.plot_prob.setZValue(10)
self.plot_prob.setXLink(self.ploti)
self.update_views()
self.ploti.vb.sigResized.connect(self.update_views)
self.plot_prob.setRange(yRange=[0, 1])
self.inline_graph_report()
def disable_mouse(plot):
plot.setMouseEnabled(False, False)
plot.setMenuEnabled(False)
disable_mouse(self.plot)
disable_mouse(self.plot_prob)
self.tooltip_items = []
self.plot.scene().installEventFilter(
HelpEventDelegate(self.help_event, self))
pen = QtGui.QPen(self.palette().color(QtGui.QPalette.Text))
for axis in ("left", "bottom"):
self.ploti.getAxis(axis).setPen(pen)
self._legend = LegendItem()
self._legend.setParentItem(self.plot)
self._legend.hide()
self._legend.anchor((1, 0), (1, 0))
def update_views(self):
self.plot_prob.setGeometry(self.plot.sceneBoundingRect())
self.plot_prob.linkedViewChanged(self.plot, self.plot_prob.XAxis)
def set_data(self, data):
self.closeContext()
self.clear()
self.data = data
if self.data is not None:
domain = self.data.domain
self.varmodel[:] = list(domain)
self.groupvarview.clear()
self.groupvarmodel = \
["(None)"] + [var for var in domain if var.is_discrete]
self.groupvarview.addItem("(None)")
for var in self.groupvarmodel[1:]:
self.groupvarview.addItem(self.icons[var], var.name)
if domain.has_discrete_class:
self.groupvar_idx = \
self.groupvarmodel[1:].index(domain.class_var) + 1
self.openContext(domain)
self.variable_idx = min(max(self.variable_idx, 0),
len(self.varmodel) - 1)
self.groupvar_idx = min(max(self.groupvar_idx, 0),
len(self.groupvarmodel) - 1)
itemmodels.select_row(self.varview, self.variable_idx)
self._setup()
def clear(self):
self.plot.clear()
self.plot_prob.clear()
self.varmodel[:] = []
self.groupvarmodel = []
self.variable_idx = -1
self.groupvar_idx = 0
self._legend.clear()
self._legend.hide()
def _setup_smoothing(self):
if not self.disc_cont and self.var and self.var.is_continuous:
self.cb_disc_cont.setText("Bin continuous variables")
self.l_smoothing_l.setText("Smooth")
self.l_smoothing_r.setText("Precise")
else:
self.cb_disc_cont.setText(
"Bin continuous variables into {} bins".format(
self.bins[self.smoothing_index]))
self.l_smoothing_l.setText(" " + str(self.bins[0]))
self.l_smoothing_r.setText(" " + str(self.bins[-1]))
def _setup(self):
self.plot.clear()
self.plot_prob.clear()
self._legend.clear()
self._legend.hide()
varidx = self.variable_idx
self.var = self.cvar = None
if varidx >= 0:
self.var = self.varmodel[varidx]
if self.groupvar_idx > 0:
self.cvar = self.groupvarmodel[self.groupvar_idx]
self.cb_prob.clear()
self.cb_prob.addItem("(None)")
self.cb_prob.addItems(self.cvar.values)
self.cb_prob.addItem("(All)")
self.show_prob = min(max(self.show_prob, 0),
len(self.cvar.values) + 1)
data = self.data
self._setup_smoothing()
if self.var is None:
return
if self.disc_cont:
data = self.data[:,
(self.var, self.cvar) if self.cvar else self.var]
disc = Orange.preprocess.discretize.EqualWidth(
n=self.bins[self.smoothing_index])
data = Orange.preprocess.Discretize(data,
method=disc,
remove_const=False)
self.var = data.domain[0]
self.set_left_axis_name()
self.enable_disable_rel_freq()
if self.cvar:
self.contingencies = \
contingency.get_contingency(data, self.var, self.cvar)
self.display_contingency()
else:
self.distributions = \
distribution.get_distribution(data, self.var)
self.display_distribution()
self.plot.autoRange()
def help_event(self, ev):
in_graph_coor = self.plot.mapSceneToView(ev.scenePos())
ctooltip = []
for vb, item in self.tooltip_items:
if isinstance(item,
pg.PlotCurveItem) and item.mouseShape().contains(
vb.mapSceneToView(ev.scenePos())):
ctooltip.append(item.tooltip)
elif isinstance(
item,
DistributionBarItem) and item.boundingRect().contains(
vb.mapSceneToView(ev.scenePos())):
ctooltip.append(item.tooltip)
if ctooltip:
QToolTip.showText(ev.screenPos(),
"\n\n".join(ctooltip),
widget=self.plotview)
return True
return False
def display_distribution(self):
dist = self.distributions
var = self.var
assert len(dist) > 0
self.plot.clear()
self.plot_prob.clear()
self.ploti.hideAxis('right')
self.tooltip_items = []
bottomaxis = self.ploti.getAxis("bottom")
bottomaxis.setLabel(var.name)
bottomaxis.resizeEvent()
self.set_left_axis_name()
if var and var.is_continuous:
bottomaxis.setTicks(None)
if not len(dist[0]):
return
edges, curve = ash_curve(
dist,
None,
m=OWDistributions.ASH_HIST,
smoothing_factor=self.smoothing_facs[self.smoothing_index])
edges = edges + (edges[1] - edges[0]) / 2
edges = edges[:-1]
item = pg.PlotCurveItem()
pen = QtGui.QPen(QtGui.QBrush(Qt.white), 3)
pen.setCosmetic(True)
item.setData(edges,
curve,
antialias=True,
stepMode=False,
fillLevel=0,
brush=QtGui.QBrush(Qt.gray),
pen=pen)
self.plot.addItem(item)
item.tooltip = "Density"
self.tooltip_items.append((self.plot, item))
else:
bottomaxis.setTicks([list(enumerate(var.values))])
for i, w in enumerate(dist):
geom = QtCore.QRectF(i - 0.33, 0, 0.66, w)
item = DistributionBarItem(geom, [1.0],
[QtGui.QColor(128, 128, 128)])
self.plot.addItem(item)
item.tooltip = "Frequency for %s: %r" % (var.values[i], w)
self.tooltip_items.append((self.plot, item))
def _on_relative_freq_changed(self):
self.set_left_axis_name()
if self.cvar and self.cvar.is_discrete:
self.display_contingency()
else:
self.display_distribution()
self.plot.autoRange()
def display_contingency(self):
"""
Set the contingency to display.
"""
cont = self.contingencies
var, cvar = self.var, self.cvar
assert len(cont) > 0
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 = [QtGui.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_facs[
self.smoothing_index]))
weights = numpy.array(weights)
sumw = numpy.sum(weights)
weights /= sumw
colors = cols
curves = [(X, Y * w) for (X, Y), w in zip(curves, weights)]
ncval = len(cvar_values)
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"]:
for (X, Y), color, w, cval in reversed(
list(zip(curvesline, colors, weights, cvar_values))):
item = pg.PlotCurveItem()
pen = QtGui.QPen(QtGui.QBrush(color), 3)
pen.setCosmetic(True)
color = QtGui.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=QtGui.QBrush(color),
pen=pen)
self.plot.addItem(item)
if t == "line":
item.tooltip = ("Normalized density " if self.relative_freq else "Density ") \
+ "\n"+ cvar.name + "=" + cval
self.tooltip_items.append((self.plot, item))
if self.show_prob:
M_EST = 5 #for M estimate
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)
# allcorrection = M_EST/sumw*numpy.sum(sumprob)/len(inter_X)
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 = QtGui.QPen(QtGui.QBrush(color),
3,
style=QtCore.Qt.DotLine)
pen.setCosmetic(True)
#prob = (Y+allcorrection/ncval)/(sumprob+allcorrection)
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)
ncval = len(cvar_values)
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 = QtCore.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 = ("Normalized frequency " if self.relative_freq else "Frequency ") \
+ "(" + cvar.name + "=" + value + "):" \
+ "\n" + 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()
bar = pg.ErrorBarItem()
pen = QtGui.QPen(QtGui.QBrush(QtGui.QColor(0)), 1)
pen.setCosmetic(True)
bar.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=QtGui.QColor(1),
pen=pen)
mark.setData([i + position], [prob],
antialias=True,
symbol="o",
fillLevel=None,
pxMode=True,
size=10,
brush=QtGui.QColor(colors[ic]),
pen=pen)
self.plot_prob.addItem(bar)
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()
def set_left_axis_name(self):
leftaxis = self.ploti.getAxis("left")
set_label = leftaxis.setLabel
if self.var and self.var.is_continuous:
set_label(["Density", "Relative density"][self.cvar is not None
and self.relative_freq])
else:
set_label(["Frequency",
"Relative frequency"][self.cvar is not None
and self.relative_freq])
leftaxis.resizeEvent()
def enable_disable_rel_freq(self):
self.cb_prob.setDisabled(self.var is None or self.cvar is None)
self.cb_rel_freq.setDisabled(self.var is None or self.cvar is None)
def _on_variable_idx_changed(self):
self.variable_idx = selected_index(self.varview)
self._setup()
def _on_groupvar_idx_changed(self):
self._setup()
def _on_set_smoothing(self):
self._setup()
def onDeleteWidget(self):
self.plot.clear()
super().onDeleteWidget()
def get_widget_name_extension(self):
if self.variable_idx >= 0:
return self.varmodel[self.variable_idx]
def send_report(self):
if self.variable_idx < 0:
return
self.report_plot()
text = "Distribution of '{}'".format(self.varmodel[self.variable_idx])
if self.groupvar_idx:
group_var = self.groupvarmodel[self.groupvar_idx]
prob = self.cb_prob
indiv_probs = 0 < prob.currentIndex() < prob.count() - 1
if not indiv_probs or self.relative_freq:
text += " grouped by '{}'".format(group_var)
if self.relative_freq:
text += " (relative frequencies)"
if indiv_probs:
text += "; probabilites for '{}={}'".format(
group_var, prob.currentText())
self.report_caption(text)
def __init__(self):
super().__init__()
self.data = None
self.distributions = None
self.contingencies = None
self.var = self.cvar = None
varbox = gui.vBox(self.controlArea, "Variable")
self.varmodel = itemmodels.VariableListModel()
self.groupvarmodel = []
self.varview = QListView(
selectionMode=QListView.SingleSelection)
self.varview.setSizePolicy(
QSizePolicy.Minimum, QSizePolicy.Expanding)
self.varview.setModel(self.varmodel)
self.varview.setSelectionModel(
itemmodels.ListSingleSelectionModel(self.varmodel))
self.varview.selectionModel().selectionChanged.connect(
self._on_variable_idx_changed)
varbox.layout().addWidget(self.varview)
box = gui.vBox(self.controlArea, "Precision")
gui.separator(self.controlArea, 4, 4)
box2 = gui.hBox(box)
self.l_smoothing_l = gui.widgetLabel(box2, "Smooth")
gui.hSlider(box2, self, "smoothing_index",
minValue=0, maxValue=len(self.smoothing_facs) - 1,
callback=self._on_set_smoothing, createLabel=False)
self.l_smoothing_r = gui.widgetLabel(box2, "Precise")
self.cb_disc_cont = gui.checkBox(
gui.indentedBox(box, sep=4),
self, "disc_cont", "Bin numeric variables",
callback=self._on_groupvar_idx_changed,
tooltip="Show numeric variables as categorical.")
box = gui.vBox(self.controlArea, "Group by")
self.icons = gui.attributeIconDict
self.groupvarview = gui.comboBox(
box, self, "groupvar_idx",
callback=self._on_groupvar_idx_changed,
valueType=str, contentsLength=12)
box2 = gui.indentedBox(box, sep=4)
self.cb_rel_freq = gui.checkBox(
box2, self, "relative_freq", "Show relative frequencies",
callback=self._on_relative_freq_changed,
tooltip="Normalize probabilities so that probabilities "
"for each group-by value sum to 1.")
gui.separator(box2)
self.cb_prob = gui.comboBox(
box2, self, "show_prob", label="Show probabilities:",
orientation=Qt.Horizontal,
callback=self._on_relative_freq_changed,
tooltip="Show probabilities for a chosen group-by value "
"(at each point probabilities for all group-by values sum to 1).")
self.plotview = pg.PlotWidget(background=None)
self.plotview.setRenderHint(QPainter.Antialiasing)
self.mainArea.layout().addWidget(self.plotview)
w = QLabel()
w.setSizePolicy(QSizePolicy.Expanding, QSizePolicy.Fixed)
self.mainArea.layout().addWidget(w, Qt.AlignCenter)
self.ploti = pg.PlotItem()
self.plot = self.ploti.vb
self.ploti.hideButtons()
self.plotview.setCentralItem(self.ploti)
self.plot_prob = pg.ViewBox()
self.ploti.hideAxis('right')
self.ploti.scene().addItem(self.plot_prob)
self.ploti.getAxis("right").linkToView(self.plot_prob)
self.ploti.getAxis("right").setLabel("Probability")
self.plot_prob.setZValue(10)
self.plot_prob.setXLink(self.ploti)
self.update_views()
self.ploti.vb.sigResized.connect(self.update_views)
self.plot_prob.setRange(yRange=[0, 1])
def disable_mouse(plot):
plot.setMouseEnabled(False, False)
plot.setMenuEnabled(False)
disable_mouse(self.plot)
disable_mouse(self.plot_prob)
self.tooltip_items = []
self.plot.scene().installEventFilter(
HelpEventDelegate(self.help_event, self))
pen = QPen(self.palette().color(QPalette.Text))
for axis in ("left", "bottom"):
self.ploti.getAxis(axis).setPen(pen)
self._legend = LegendItem()
self._legend.setParentItem(self.plot)
self._legend.hide()
self._legend.anchor((1, 0), (1, 0))
class OWMDS(widget.OWWidget):
name = "MDS"
description = "Two-dimensional data projection by multidimensional " \
"scaling constructed from a distance matrix."
icon = "icons/MDS.svg"
inputs = [("Data", Orange.data.Table, "set_data"),
("Distances", Orange.misc.DistMatrix, "set_disimilarity")]
outputs = [("Data", Orange.data.Table, widget.Default),
("Data Subset", Orange.data.Table)]
#: Initialization type
PCA, Random = 0, 1
#: Refresh rate
RefreshRate = [
("Every iteration", 1),
("Every 5 steps", 5),
("Every 10 steps", 10),
("Every 25 steps", 25),
("Every 50 steps", 50),
("None", -1)
]
#: Runtime state
Running, Finished, Waiting = 1, 2, 3
settingsHandler = settings.DomainContextHandler()
max_iter = settings.Setting(300)
initialization = settings.Setting(PCA)
refresh_rate = settings.Setting(3)
# output embedding role.
NoRole, AttrRole, MetaRole = 0, 1, 2
output_embedding_role = settings.Setting(1)
autocommit = settings.Setting(True)
color_index = settings.ContextSetting(0, not_attribute=True)
shape_index = settings.ContextSetting(0, not_attribute=True)
size_index = settings.ContextSetting(0, not_attribute=True)
label_index = settings.ContextSetting(0, not_attribute=True)
symbol_size = settings.Setting(8)
symbol_opacity = settings.Setting(230)
legend_anchor = settings.Setting(((1, 0), (1, 0)))
def __init__(self, parent=None):
super().__init__(parent)
self.matrix = None
self.data = None
self.matrix_data = None
self.signal_data = None
self._pen_data = None
self._shape_data = None
self._size_data = None
self._label_data = None
self._scatter_item = None
self._legend_item = None
self._selection_mask = None
self._invalidated = False
self._effective_matrix = None
self.__update_loop = None
self.__state = OWMDS.Waiting
self.__in_next_step = False
box = gui.widgetBox(self.controlArea, "MDS Optimization")
form = QtGui.QFormLayout(
labelAlignment=Qt.AlignLeft,
formAlignment=Qt.AlignLeft,
fieldGrowthPolicy=QtGui.QFormLayout.AllNonFixedFieldsGrow,
)
form.addRow("Max iterations:",
gui.spin(box, self, "max_iter", 10, 10 ** 4, step=1))
form.addRow("Initialization",
gui.comboBox(box, self, "initialization",
items=["PCA (Torgerson)", "Random"],
callback=self.__invalidate_embedding))
box.layout().addLayout(form)
form.addRow("Refresh",
gui.comboBox(
box, self, "refresh_rate",
items=[t for t, _ in OWMDS.RefreshRate],
callback=self.__invalidate_refresh))
self.runbutton = gui.button(
box, self, "Run", callback=self._toggle_run)
box = gui.widgetBox(self.controlArea, "Graph")
self.colorvar_model = itemmodels.VariableListModel()
cb = gui.comboBox(box, self, "color_index", box="Color",
callback=self._on_color_index_changed)
cb.setModel(self.colorvar_model)
cb.box.setFlat(True)
self.shapevar_model = itemmodels.VariableListModel()
cb = gui.comboBox(box, self, "shape_index", box="Shape",
callback=self._on_shape_index_changed)
cb.setModel(self.shapevar_model)
cb.box.setFlat(True)
self.sizevar_model = itemmodels.VariableListModel()
cb = gui.comboBox(box, self, "size_index", "Size",
callback=self._on_size_index_changed)
cb.setModel(self.sizevar_model)
cb.box.setFlat(True)
self.labelvar_model = itemmodels.VariableListModel()
cb = gui.comboBox(box, self, "label_index", "Label",
callback=self._on_label_index_changed)
cb.setModel(self.labelvar_model)
cb.box.setFlat(True)
form = QtGui.QFormLayout(
labelAlignment=Qt.AlignLeft,
formAlignment=Qt.AlignLeft,
fieldGrowthPolicy=QtGui.QFormLayout.AllNonFixedFieldsGrow,
)
form.addRow("Symbol size",
gui.hSlider(box, self, "symbol_size",
minValue=1, maxValue=20,
callback=self._on_size_index_changed,
createLabel=False))
form.addRow("Symbol opacity",
gui.hSlider(box, self, "symbol_opacity",
minValue=100, maxValue=255, step=100,
callback=self._on_color_index_changed,
createLabel=False))
box.layout().addLayout(form)
box = QtGui.QGroupBox("Zoom/Select", )
box.setLayout(QtGui.QHBoxLayout())
group = QtGui.QActionGroup(self, exclusive=True)
def icon(name):
path = "icons/Dlg_{}.png".format(name)
path = pkg_resources.resource_filename(widget.__name__, path)
return QtGui.QIcon(path)
action_select = QtGui.QAction(
"Select", self, checkable=True, checked=True, icon=icon("arrow"),
shortcut=QtGui.QKeySequence(Qt.ControlModifier + Qt.Key_1))
action_zoom = QtGui.QAction(
"Zoom", self, checkable=True, checked=False, icon=icon("zoom"),
shortcut=QtGui.QKeySequence(Qt.ControlModifier + Qt.Key_2))
action_pan = QtGui.QAction(
"Pan", self, checkable=True, checked=False, icon=icon("pan_hand"),
shortcut=QtGui.QKeySequence(Qt.ControlModifier + Qt.Key_3))
action_reset_zoom = QtGui.QAction(
"Zoom to fit", self, icon=icon("zoom_reset"),
shortcut=QtGui.QKeySequence(Qt.ControlModifier + Qt.Key_0))
action_reset_zoom.triggered.connect(
lambda: self.plot.autoRange())
group.addAction(action_select)
group.addAction(action_zoom)
group.addAction(action_pan)
self.addActions(group.actions() + [action_reset_zoom])
action_select.setChecked(True)
def button(action):
b = QtGui.QToolButton()
b.setToolButtonStyle(Qt.ToolButtonIconOnly)
b.setDefaultAction(action)
return b
box.layout().addWidget(button(action_select))
box.layout().addWidget(button(action_zoom))
box.layout().addWidget(button(action_pan))
box.layout().addSpacing(4)
box.layout().addWidget(button(action_reset_zoom))
box.layout().addStretch()
self.controlArea.layout().addWidget(box)
gui.rubber(self.controlArea)
box = gui.widgetBox(self.controlArea, "Output")
cb = gui.comboBox(box, self, "output_embedding_role",
box="Append coordinates",
items=["Do not append", "As attributes", "As metas"],
callback=self._invalidate_output)
cb.box.setFlat(True)
gui.auto_commit(box, self, "autocommit", "Send data",
checkbox_label="Send after any change",
box=None)
self.plot = pg.PlotWidget(background="w", enableMenu=False)
self.mainArea.layout().addWidget(self.plot)
self.selection_tool = PlotSelectionTool(
parent=self, selectionMode=PlotSelectionTool.Lasso)
self.zoom_tool = PlotZoomTool(parent=self)
self.pan_tool = PlotPanTool(parent=self)
self.pinch_tool = PlotPinchZoomTool(parent=self)
self.pinch_tool.setViewBox(self.plot.getViewBox())
self.selection_tool.setViewBox(self.plot.getViewBox())
self.selection_tool.selectionFinished.connect(self.__selection_end)
self.current_tool = self.selection_tool
def activate_tool(action):
self.current_tool.setViewBox(None)
if action is action_select:
active, cur = self.selection_tool, Qt.ArrowCursor
elif action is action_zoom:
active, cur = self.zoom_tool, Qt.ArrowCursor
elif action is action_pan:
active, cur = self.pan_tool, Qt.OpenHandCursor
self.current_tool = active
self.current_tool.setViewBox(self.plot.getViewBox())
self.plot.getViewBox().setCursor(QtGui.QCursor(cur))
group.triggered[QtGui.QAction].connect(activate_tool)
def set_data(self, data):
self.signal_data = data
if self.matrix and data is not None and len(self.matrix.X) == len(data):
self.closeContext()
self.data = data
self.update_controls()
self.openContext(data)
else:
self._invalidated = True
self._selection_mask = None
def set_disimilarity(self, matrix):
self.matrix = matrix
if matrix and matrix.row_items:
self.matrix_data = matrix.row_items
if matrix is None:
self.matrix_data = None
self._invalidated = True
self._selection_mask = None
def _clear(self):
self._pen_data = None
self._shape_data = None
self._size_data = None
self._label_data = None
self.colorvar_model[:] = ["Same color"]
self.shapevar_model[:] = ["Same shape"]
self.sizevar_model[:] = ["Same size"]
self.labelvar_model[:] = ["No labels"]
self.color_index = 0
self.shape_index = 0
self.size_index = 0
self.label_index = 0
self.__set_update_loop(None)
self.__state = OWMDS.Waiting
def _clear_plot(self):
self.plot.clear()
self._scatter_item = None
if self._legend_item is not None:
anchor = legend_anchor_pos(self._legend_item)
if anchor is not None:
self.legend_anchor = anchor
if self._legend_item.scene() is not None:
self._legend_item.scene().removeItem(self._legend_item)
self._legend_item = None
def update_controls(self):
if getattr(self.matrix, 'axis', 1) == 0:
# Column-wise distances
attr = "Attribute names"
self.labelvar_model[:] = ["No labels", attr]
self.shapevar_model[:] = ["Same shape", attr]
self.colorvar_model[:] = ["Same color", attr]
self.color_index = list(self.colorvar_model).index(attr)
self.shape_index = list(self.shapevar_model).index(attr)
else:
# initialize the graph state from data
domain = self.data.domain
all_vars = list(domain.variables + domain.metas)
cd_vars = [var for var in all_vars if var.is_primitive()]
disc_vars = [var for var in all_vars if var.is_discrete]
cont_vars = [var for var in all_vars if var.is_continuous]
str_vars = [var for var in all_vars
if var.is_discrete or var.is_string]
self.colorvar_model[:] = chain(["Same color"],
[self.colorvar_model.Separator],
cd_vars)
self.shapevar_model[:] = chain(["Same shape"],
[self.shapevar_model.Separator],
disc_vars)
self.sizevar_model[:] = chain(["Same size", "Stress"],
[self.sizevar_model.Separator],
cont_vars)
self.labelvar_model[:] = chain(["No labels"],
[self.labelvar_model.Separator],
str_vars)
if domain.class_var is not None:
self.color_index = list(self.colorvar_model).index(domain.class_var)
def _initialize(self):
# clear everything
self.closeContext()
self._clear()
self.data = None
self._effective_matrix = None
self.embedding = None
# if no data nor matrix is present reset plot
if self.signal_data is None and self.matrix is None:
self._update_plot()
return
if self.signal_data and self.matrix_data and len(self.signal_data) != len(self.matrix_data):
self.error(1, "Data and distances dimensions do not match.")
self._update_plot()
return
self.error(1)
if self.signal_data:
self.data = self.signal_data
elif self.matrix_data:
self.data = self.matrix_data
if self.matrix:
self._effective_matrix = self.matrix
if self.matrix.axis == 0:
self.data = None
else:
self._effective_matrix = Orange.distance.Euclidean(self.data)
self.update_controls()
self.openContext(self.data)
def _toggle_run(self):
if self.__state == OWMDS.Running:
self.stop()
self._invalidate_output()
else:
self.start()
def start(self):
if self.__state == OWMDS.Running:
return
elif self.__state == OWMDS.Finished:
# Resume/continue from a previous run
self.__start()
elif self.__state == OWMDS.Waiting and \
self._effective_matrix is not None:
self.__start()
def stop(self):
if self.__state == OWMDS.Running:
self.__set_update_loop(None)
def __start(self):
X = self._effective_matrix.X
if self.embedding is not None:
init = self.embedding
elif self.initialization == OWMDS.PCA:
init = torgerson(X, n_components=2)
else:
init = None
# number of iterations per single GUI update step
_, step_size = OWMDS.RefreshRate[self.refresh_rate]
if step_size == -1:
step_size = self.max_iter
def update_loop(X, max_iter, step, init):
"""
return an iterator over successive improved MDS point embeddings.
"""
# NOTE: this code MUST NOT call into QApplication.processEvents
done = False
iterations_done = 0
oldstress = numpy.finfo(numpy.float).max
while not done:
step_iter = min(max_iter - iterations_done, step)
mds = Orange.projection.MDS(
dissimilarity="precomputed", n_components=2,
n_init=1, max_iter=step_iter)
mdsfit = mds.fit(X, init=init)
iterations_done += step_iter
embedding, stress = mdsfit.embedding_, mdsfit.stress_
stress /= numpy.sqrt(numpy.sum(embedding ** 2, axis=1)).sum()
if iterations_done >= max_iter:
done = True
elif (oldstress - stress) < mds.params["eps"]:
done = True
init = embedding
oldstress = stress
yield embedding, mdsfit.stress_, iterations_done / max_iter
self.__set_update_loop(update_loop(X, self.max_iter, step_size, init))
self.progressBarInit(processEvents=None)
def __set_update_loop(self, loop):
"""
Set the update `loop` coroutine.
The `loop` is a generator yielding `(embedding, stress, progress)`
tuples where `embedding` is a `(N, 2) ndarray` of current updated
MDS points, `stress` is the current stress and `progress` a float
ratio (0 <= progress <= 1)
If an existing update loop is already in palace it is interrupted
(closed).
.. note::
The `loop` must not explicitly yield control flow to the event
loop (i.e. call `QApplication.proceesEvents`)
"""
if self.__update_loop is not None:
self.__update_loop.close()
self.__update_loop = None
self.progressBarFinished(processEvents=None)
self.__update_loop = loop
if loop is not None:
self.progressBarInit(processEvents=None)
self.setStatusMessage("Running")
self.runbutton.setText("Stop")
self.__state = OWMDS.Running
QtGui.QApplication.postEvent(self, QEvent(QEvent.User))
else:
self.setStatusMessage("")
self.runbutton.setText("Start")
self.__state = OWMDS.Finished
def __next_step(self):
if self.__update_loop is None:
return
loop = self.__update_loop
try:
embedding, stress, progress = next(self.__update_loop)
assert self.__update_loop is loop
except StopIteration:
self.__set_update_loop(None)
self.unconditional_commit()
else:
self.progressBarSet(100.0 * progress, processEvents=None)
self.embedding = embedding
self._update_plot()
# schedule next update
QtGui.QApplication.postEvent(
self, QEvent(QEvent.User), Qt.LowEventPriority)
def customEvent(self, event):
if event.type() == QEvent.User and self.__update_loop is not None:
if not self.__in_next_step:
self.__in_next_step = True
try:
self.__next_step()
finally:
self.__in_next_step = False
else:
warnings.warn(
"Re-entry in update loop detected. "
"A rogue `proccessEvents` is on the loose.",
RuntimeWarning)
# re-schedule the update iteration.
QtGui.QApplication.postEvent(self, QEvent(QEvent.User))
return super().customEvent(event)
def __invalidate_embedding(self):
state = self.__state
if self.__update_loop is not None:
self.__set_update_loop(None)
X = self._effective_matrix.X
if self.initialization == OWMDS.PCA:
self.embedding = torgerson(X)
else:
self.embedding = numpy.random.rand(len(X), 2)
self._update_plot()
# restart the optimization if it was interrupted.
if state == OWMDS.Running:
self.__start()
def __invalidate_refresh(self):
state = self.__state
if self.__update_loop is not None:
self.__set_update_loop(None)
# restart the optimization if it was interrupted.
# TODO: decrease the max iteration count by the already
# completed iterations count.
if state == OWMDS.Running:
self.__start()
def handleNewSignals(self):
if self._invalidated:
self._invalidated = False
self._initialize()
self.start()
self._update_plot()
self.unconditional_commit()
def _invalidate_output(self):
self.commit()
def _on_color_index_changed(self):
self._pen_data = None
self._update_plot()
def _on_shape_index_changed(self):
self._shape_data = None
self._update_plot()
def _on_size_index_changed(self):
self._size_data = None
self._update_plot()
def _on_label_index_changed(self):
self._label_data = None
self._update_plot()
def _update_plot(self):
self._clear_plot()
if self.embedding is not None:
self._setup_plot()
def _setup_plot(self):
have_data = self.data is not None
have_matrix_transposed = self.matrix is not None and not self.matrix.axis
def column(data, variable):
a, _ = data.get_column_view(variable)
return a.ravel()
def attributes(matrix):
return matrix.row_items.domain.attributes
def scale(a):
dmin, dmax = numpy.nanmin(a), numpy.nanmax(a)
if dmax - dmin > 0:
return (a - dmin) / (dmax - dmin)
else:
return numpy.zeros_like(a)
if self._pen_data is None:
if self._selection_mask is not None:
pointflags = numpy.where(
self._selection_mask,
mdsplotutils.Selected, mdsplotutils.NoFlags)
else:
pointflags = None
if have_data and self.color_index > 0:
color_var = self.colorvar_model[self.color_index]
if color_var.is_discrete:
palette = colorpalette.ColorPaletteGenerator(
len(color_var.values)
)
else:
palette = None
color_data = mdsplotutils.color_data(
self.data, color_var, plotstyle=mdsplotutils.plotstyle)
color_data = numpy.hstack(
(color_data,
numpy.full((len(color_data), 1), self.symbol_opacity))
)
pen_data = mdsplotutils.pen_data(color_data, pointflags)
elif have_matrix_transposed and self.colorvar_model[self.color_index] == 'Attribute names':
attr = attributes(self.matrix)
palette = colorpalette.ColorPaletteGenerator(len(attr))
color_data = [palette.getRGB(i) for i in range(len(attr))]
color_data = numpy.hstack(
color_data,
numpy.full((len(color_data), 1), self.symbol_opacity)
)
pen_data = mdsplotutils.pen_data(color_data, pointflags)
else:
pen_data = make_pen(QtGui.QColor(Qt.darkGray), cosmetic=True)
pen_data = numpy.full(len(self.data), pen_data, dtype=object)
self._pen_data = pen_data
if self._shape_data is None:
if have_data and self.shape_index > 0:
Symbols = ScatterPlotItem.Symbols
symbols = numpy.array(list(Symbols.keys()))
shape_var = self.shapevar_model[self.shape_index]
data = column(self.data, shape_var)
data = data % (len(Symbols) - 1)
data[numpy.isnan(data)] = len(Symbols) - 1
shape_data = symbols[data.astype(int)]
elif have_matrix_transposed and self.shapevar_model[self.shape_index] == 'Attribute names':
Symbols = ScatterPlotItem.Symbols
symbols = numpy.array(list(Symbols.keys()))
attr = [i % (len(Symbols) - 1) for i, _ in enumerate(attributes(self.matrix))]
shape_data = symbols[attr]
else:
shape_data = "o"
self._shape_data = shape_data
if self._size_data is None:
MinPointSize = 3
point_size = self.symbol_size + MinPointSize
if have_data and self.size_index == 1:
# size by stress
size_data = stress(self.embedding, self._effective_matrix.X)
size_data = scale(size_data)
size_data = MinPointSize + size_data * point_size
elif have_data and self.size_index > 0:
size_var = self.sizevar_model[self.size_index]
size_data = column(self.data, size_var)
size_data = scale(size_data)
size_data = MinPointSize + size_data * point_size
else:
size_data = point_size
if self._label_data is None:
if have_data and self.label_index > 0:
label_var = self.labelvar_model[self.label_index]
label_data = column(self.data, label_var)
label_data = [label_var.repr_val(val) for val in label_data]
label_items = [pg.TextItem(text, anchor=(0.5, 0))
for text in label_data]
elif have_matrix_transposed and self.labelvar_model[self.label_index] == 'Attribute names':
attr = attributes(self.matrix)
label_items = [pg.TextItem(str(text), anchor=(0.5, 0))
for text in attr]
else:
label_items = None
self._label_data = label_items
self._scatter_item = item = ScatterPlotItem(
x=self.embedding[:, 0], y=self.embedding[:, 1],
pen=self._pen_data, symbol=self._shape_data,
brush=QtGui.QBrush(Qt.transparent),
size=size_data, data=numpy.arange(len(self.data)),
antialias=True
)
self.plot.addItem(item)
if self._label_data is not None:
for (x, y), text_item in zip(self.embedding, self._label_data):
self.plot.addItem(text_item)
text_item.setPos(x, y)
self._legend_item = LegendItem()
self._legend_item.setParentItem(self.plot.getViewBox())
self._legend_item.anchor(*self.legend_anchor)
color_var = shape_var = None
if have_data and 1 <= self.color_index < len(self.colorvar_model):
color_var = self.colorvar_model[self.color_index]
assert isinstance(color_var, Orange.data.Variable)
if have_data and 1 <= self.shape_index < len(self.shapevar_model):
shape_var = self.shapevar_model[self.shape_index]
assert isinstance(shape_var, Orange.data.Variable)
if shape_var is not None or \
(color_var is not None and color_var.is_discrete):
legend_data = mdsplotutils.legend_data(
color_var, shape_var, plotstyle=mdsplotutils.plotstyle)
for color, symbol, text in legend_data:
self._legend_item.addItem(
ScatterPlotItem(pen=color, brush=color, symbol=symbol,
size=10),
text
)
else:
self._legend_item.hide()
def commit(self):
if self.embedding is not None:
output = embedding = Orange.data.Table.from_numpy(
Orange.data.Domain([Orange.data.ContinuousVariable("X"),
Orange.data.ContinuousVariable("Y")]),
self.embedding
)
else:
output = embedding = None
if self.embedding is not None and self.data is not None:
domain = self.data.domain
attrs = domain.attributes
class_vars = domain.class_vars
metas = domain.metas
if self.output_embedding_role == OWMDS.AttrRole:
attrs = attrs + embedding.domain.attributes
elif self.output_embedding_role == OWMDS.MetaRole:
metas = metas + embedding.domain.attributes
domain = Orange.data.Domain(attrs, class_vars, metas)
output = Orange.data.Table.from_table(domain, self.data)
if self.output_embedding_role == OWMDS.AttrRole:
output.X[:, -2:] = embedding.X
elif self.output_embedding_role == OWMDS.MetaRole:
output.metas[:, -2:] = embedding.X
self.send("Data", output)
if output is not None and self._selection_mask is not None and \
numpy.any(self._selection_mask):
subset = output[self._selection_mask]
else:
subset = None
self.send("Data Subset", subset)
def onDeleteWidget(self):
super().onDeleteWidget()
self._clear_plot()
self._clear()
def __selection_end(self, path):
self.select(path)
self._pen_data = None
self._update_plot()
self._invalidate_output()
def select(self, region):
item = self._scatter_item
if item is None:
return
indices = numpy.array(
[spot.data() for spot in item.points()
if region.contains(spot.pos())],
dtype=int)
if not QtGui.QApplication.keyboardModifiers() & Qt.ControlModifier:
self._selection_mask = None
self.select_indices(indices)
def select_indices(self, indices):
if self.data is None:
return
if self._selection_mask is None:
self._selection_mask = numpy.zeros(len(self.data), dtype=bool)
self._selection_mask[indices] = True
def __init__(self):
super().__init__()
self.data = None
self.distributions = None
self.contingencies = None
self.var = self.cvar = None
varbox = gui.widgetBox(self.controlArea, "Variable")
self.varmodel = itemmodels.VariableListModel()
self.groupvarmodel = []
self.varview = QtGui.QListView(
selectionMode=QtGui.QListView.SingleSelection)
self.varview.setSizePolicy(QtGui.QSizePolicy.Minimum,
QtGui.QSizePolicy.Expanding)
self.varview.setModel(self.varmodel)
self.varview.setSelectionModel(
itemmodels.ListSingleSelectionModel(self.varmodel))
self.varview.selectionModel().selectionChanged.connect(
self._on_variable_idx_changed)
varbox.layout().addWidget(self.varview)
box = gui.widgetBox(self.controlArea, "Precision")
gui.separator(self.controlArea, 4, 4)
box2 = gui.widgetBox(box, orientation="horizontal")
self.l_smoothing_l = gui.widgetLabel(box2, "Smooth")
gui.hSlider(box2,
self,
"smoothing_index",
minValue=0,
maxValue=len(self.smoothing_facs) - 1,
callback=self._on_set_smoothing,
createLabel=False)
self.l_smoothing_r = gui.widgetLabel(box2, "Precise")
self.cb_disc_cont = gui.checkBox(
gui.indentedBox(box, sep=4),
self,
"disc_cont",
"Bin continuous variables",
callback=self._on_groupvar_idx_changed,
tooltip="Show continuous variables as discrete.")
box = gui.widgetBox(self.controlArea, "Group by")
self.icons = gui.attributeIconDict
self.groupvarview = gui.comboBox(
box,
self,
"groupvar_idx",
callback=self._on_groupvar_idx_changed,
valueType=str,
contentsLength=12)
box2 = gui.indentedBox(box, sep=4)
self.cb_rel_freq = gui.checkBox(
box2,
self,
"relative_freq",
"Show relative frequencies",
callback=self._on_relative_freq_changed,
tooltip=
"Normalize probabilities so that probabilities for each group-by value sum to 1."
)
gui.separator(box2)
self.cb_prob = gui.comboBox(
box2,
self,
"show_prob",
label="Show probabilities",
orientation="horizontal",
callback=self._on_relative_freq_changed,
tooltip=
"Show probabilities for a chosen group-by value (at each point probabilities for all group-by values sum to 1)."
)
self.plotview = pg.PlotWidget(background=None)
self.plotview.setRenderHint(QtGui.QPainter.Antialiasing)
self.mainArea.layout().addWidget(self.plotview)
w = QtGui.QLabel()
w.setSizePolicy(QtGui.QSizePolicy.Expanding, QtGui.QSizePolicy.Fixed)
self.mainArea.layout().addWidget(w, Qt.AlignCenter)
self.ploti = pg.PlotItem()
self.plot = self.ploti.vb
self.ploti.hideButtons()
self.plotview.setCentralItem(self.ploti)
self.plot_prob = pg.ViewBox()
self.ploti.hideAxis('right')
self.ploti.scene().addItem(self.plot_prob)
self.ploti.getAxis("right").linkToView(self.plot_prob)
self.ploti.getAxis("right").setLabel("Probability")
self.plot_prob.setZValue(10)
self.plot_prob.setXLink(self.ploti)
self.update_views()
self.ploti.vb.sigResized.connect(self.update_views)
self.plot_prob.setRange(yRange=[0, 1])
self.inline_graph_report()
def disable_mouse(plot):
plot.setMouseEnabled(False, False)
plot.setMenuEnabled(False)
disable_mouse(self.plot)
disable_mouse(self.plot_prob)
self.tooltip_items = []
self.plot.scene().installEventFilter(
HelpEventDelegate(self.help_event, self))
pen = QtGui.QPen(self.palette().color(QtGui.QPalette.Text))
for axis in ("left", "bottom"):
self.ploti.getAxis(axis).setPen(pen)
self._legend = LegendItem()
self._legend.setParentItem(self.plot)
self._legend.hide()
self._legend.anchor((1, 0), (1, 0))
def _setup_plot(self):
have_data = self.data is not None
have_matrix_transposed = self.matrix is not None and not self.matrix.axis
def column(data, variable):
a, _ = data.get_column_view(variable)
return a.ravel()
def attributes(matrix):
return matrix.row_items.domain.attributes
def scale(a):
dmin, dmax = numpy.nanmin(a), numpy.nanmax(a)
if dmax - dmin > 0:
return (a - dmin) / (dmax - dmin)
else:
return numpy.zeros_like(a)
if self._pen_data is None:
if self._selection_mask is not None:
pointflags = numpy.where(
self._selection_mask,
mdsplotutils.Selected, mdsplotutils.NoFlags)
else:
pointflags = None
if have_data and self.color_index > 0:
color_var = self.colorvar_model[self.color_index]
if color_var.is_discrete:
palette = colorpalette.ColorPaletteGenerator(
len(color_var.values)
)
else:
palette = None
color_data = mdsplotutils.color_data(
self.data, color_var, plotstyle=mdsplotutils.plotstyle)
color_data = numpy.hstack(
(color_data,
numpy.full((len(color_data), 1), self.symbol_opacity))
)
pen_data = mdsplotutils.pen_data(color_data, pointflags)
elif have_matrix_transposed and self.colorvar_model[self.color_index] == 'Attribute names':
attr = attributes(self.matrix)
palette = colorpalette.ColorPaletteGenerator(len(attr))
color_data = [palette.getRGB(i) for i in range(len(attr))]
color_data = numpy.hstack(
color_data,
numpy.full((len(color_data), 1), self.symbol_opacity)
)
pen_data = mdsplotutils.pen_data(color_data, pointflags)
else:
pen_data = make_pen(QtGui.QColor(Qt.darkGray), cosmetic=True)
pen_data = numpy.full(len(self.data), pen_data, dtype=object)
self._pen_data = pen_data
if self._shape_data is None:
if have_data and self.shape_index > 0:
Symbols = ScatterPlotItem.Symbols
symbols = numpy.array(list(Symbols.keys()))
shape_var = self.shapevar_model[self.shape_index]
data = column(self.data, shape_var)
data = data % (len(Symbols) - 1)
data[numpy.isnan(data)] = len(Symbols) - 1
shape_data = symbols[data.astype(int)]
elif have_matrix_transposed and self.shapevar_model[self.shape_index] == 'Attribute names':
Symbols = ScatterPlotItem.Symbols
symbols = numpy.array(list(Symbols.keys()))
attr = [i % (len(Symbols) - 1) for i, _ in enumerate(attributes(self.matrix))]
shape_data = symbols[attr]
else:
shape_data = "o"
self._shape_data = shape_data
if self._size_data is None:
MinPointSize = 3
point_size = self.symbol_size + MinPointSize
if have_data and self.size_index == 1:
# size by stress
size_data = stress(self.embedding, self._effective_matrix.X)
size_data = scale(size_data)
size_data = MinPointSize + size_data * point_size
elif have_data and self.size_index > 0:
size_var = self.sizevar_model[self.size_index]
size_data = column(self.data, size_var)
size_data = scale(size_data)
size_data = MinPointSize + size_data * point_size
else:
size_data = point_size
if self._label_data is None:
if have_data and self.label_index > 0:
label_var = self.labelvar_model[self.label_index]
label_data = column(self.data, label_var)
label_data = [label_var.repr_val(val) for val in label_data]
label_items = [pg.TextItem(text, anchor=(0.5, 0))
for text in label_data]
elif have_matrix_transposed and self.labelvar_model[self.label_index] == 'Attribute names':
attr = attributes(self.matrix)
label_items = [pg.TextItem(str(text), anchor=(0.5, 0))
for text in attr]
else:
label_items = None
self._label_data = label_items
self._scatter_item = item = ScatterPlotItem(
x=self.embedding[:, 0], y=self.embedding[:, 1],
pen=self._pen_data, symbol=self._shape_data,
brush=QtGui.QBrush(Qt.transparent),
size=size_data, data=numpy.arange(len(self.data)),
antialias=True
)
self.plot.addItem(item)
if self._label_data is not None:
for (x, y), text_item in zip(self.embedding, self._label_data):
self.plot.addItem(text_item)
text_item.setPos(x, y)
self._legend_item = LegendItem()
self._legend_item.setParentItem(self.plot.getViewBox())
self._legend_item.anchor(*self.legend_anchor)
color_var = shape_var = None
if have_data and 1 <= self.color_index < len(self.colorvar_model):
color_var = self.colorvar_model[self.color_index]
assert isinstance(color_var, Orange.data.Variable)
if have_data and 1 <= self.shape_index < len(self.shapevar_model):
shape_var = self.shapevar_model[self.shape_index]
assert isinstance(shape_var, Orange.data.Variable)
if shape_var is not None or \
(color_var is not None and color_var.is_discrete):
legend_data = mdsplotutils.legend_data(
color_var, shape_var, plotstyle=mdsplotutils.plotstyle)
for color, symbol, text in legend_data:
self._legend_item.addItem(
ScatterPlotItem(pen=color, brush=color, symbol=symbol,
size=10),
text
)
else:
self._legend_item.hide()
def __init__(self):
super().__init__()
self.varmodel = itemmodels.VariableListModel()
self.groupvarmodel = []
self.distributions = [
distribution.value for distribution in self.available_distributions
]
box = gui.vBox(self.controlArea, 'Tests')
gui.radioButtonsInBox(
box,
self,
'test_idx',
btnLabels=[test.name for test in self.available_tests],
callback=self.test_changed,
)
box = gui.vBox(self.controlArea, 'Distributions')
self.distribution_choose = gui.radioButtonsInBox(
box,
self,
'distribution_idx',
btnLabels=self.distributions,
callback=self.distribution_changed,
)
self.column_chose = gui.comboBox(
self.controlArea,
self,
'column_idx',
box='Selected column',
items=[],
orientation=Qt.Horizontal,
callback=self.column_changed,
)
self.available_columns = itemmodels.VariableListModel(parent=self)
self.column_chose.setModel(self.available_columns)
self.infolabel = gui.widgetLabel(box, "<center>p-value: </center>")
self.mainArea.setMinimumWidth(800)
self.own_distribution_choose = gui.comboBox(
self.controlArea,
self,
'own_distribution_idx',
box='Own distribution',
items=[],
orientation=Qt.Horizontal,
callback=self.column_changed,
)
self.own_distribution_choose.setModel(self.available_columns)
self.data = None
self.plotview = pg.PlotWidget(background=None)
self.plotview.setRenderHint(QPainter.Antialiasing)
self.mainArea.layout().addWidget(self.plotview)
w = QLabel()
w.setSizePolicy(QSizePolicy.Expanding, QSizePolicy.Fixed)
self.mainArea.layout().addWidget(w, Qt.AlignCenter)
self.ploti = pg.PlotItem()
self.box_scene = self.ploti.vb
self.ploti.hideButtons()
self.plotview.setCentralItem(self.ploti)
self.plot_prob = pg.ViewBox()
self.ploti.scene().addItem(self.plot_prob)
self.ploti.getAxis("right").linkToView(self.plot_prob)
self.ploti.getAxis("right").setLabel("Probability")
self.plot_prob.setZValue(10)
self.plot_prob.setXLink(self.ploti)
self.update_views()
self.ploti.vb.sigResized.connect(self.update_views)
self.plot_prob.setRange(yRange=[0, 1])
def disable_mouse(box_scene):
box_scene.setMouseEnabled(False, False)
box_scene.setMenuEnabled(False)
disable_mouse(self.box_scene)
disable_mouse(self.plot_prob)
self.tooltip_items = []
pen = QPen(self.palette().color(QPalette.Text))
for axis in ("left", "bottom"):
self.ploti.getAxis(axis).setPen(pen)
self._legend = LegendItem()
self._legend.setParentItem(self.box_scene)
self._legend.hide()
self._legend.anchor((1, 0), (1, 0))
self.test_changed()
class DistributionTest(OWWidget):
name = 'Distribution Test'
description = 'Check if data is in given distribution.'
icon = 'icons/disttest.svg'
want_main_area = True
buttons_area_orientation = Qt.Vertical
resizing_enabled = True
inputs = [('Data', Orange.data.Table, 'set_data')]
available_tests = (
KolmogorovSmirnov,
AndersonDarling,
ShapiroWilk,
ChiSquare,
)
settingsHandler = settings.DomainContextHandler(
match_values=settings.DomainContextHandler.MATCH_VALUES_ALL)
#: Selected variable index
available_distributions = [d for d in Distribution]
test_idx = 0
distribution_idx = 0
column_idx = 0
own_distribution_idx = 0
relative_freq = settings.Setting(False)
smoothing_index = settings.Setting(5)
show_prob = settings.ContextSetting(0)
graph_name = "box_scene"
ASH_HIST = 50
bins = [2, 3, 4, 5, 8, 10, 12, 15, 20, 30, 50]
def __init__(self):
super().__init__()
self.varmodel = itemmodels.VariableListModel()
self.groupvarmodel = []
self.distributions = [
distribution.value for distribution in self.available_distributions
]
box = gui.vBox(self.controlArea, 'Tests')
gui.radioButtonsInBox(
box,
self,
'test_idx',
btnLabels=[test.name for test in self.available_tests],
callback=self.test_changed,
)
box = gui.vBox(self.controlArea, 'Distributions')
self.distribution_choose = gui.radioButtonsInBox(
box,
self,
'distribution_idx',
btnLabels=self.distributions,
callback=self.distribution_changed,
)
self.column_chose = gui.comboBox(
self.controlArea,
self,
'column_idx',
box='Selected column',
items=[],
orientation=Qt.Horizontal,
callback=self.column_changed,
)
self.available_columns = itemmodels.VariableListModel(parent=self)
self.column_chose.setModel(self.available_columns)
self.infolabel = gui.widgetLabel(box, "<center>p-value: </center>")
self.mainArea.setMinimumWidth(800)
self.own_distribution_choose = gui.comboBox(
self.controlArea,
self,
'own_distribution_idx',
box='Own distribution',
items=[],
orientation=Qt.Horizontal,
callback=self.column_changed,
)
self.own_distribution_choose.setModel(self.available_columns)
self.data = None
self.plotview = pg.PlotWidget(background=None)
self.plotview.setRenderHint(QPainter.Antialiasing)
self.mainArea.layout().addWidget(self.plotview)
w = QLabel()
w.setSizePolicy(QSizePolicy.Expanding, QSizePolicy.Fixed)
self.mainArea.layout().addWidget(w, Qt.AlignCenter)
self.ploti = pg.PlotItem()
self.box_scene = self.ploti.vb
self.ploti.hideButtons()
self.plotview.setCentralItem(self.ploti)
self.plot_prob = pg.ViewBox()
self.ploti.scene().addItem(self.plot_prob)
self.ploti.getAxis("right").linkToView(self.plot_prob)
self.ploti.getAxis("right").setLabel("Probability")
self.plot_prob.setZValue(10)
self.plot_prob.setXLink(self.ploti)
self.update_views()
self.ploti.vb.sigResized.connect(self.update_views)
self.plot_prob.setRange(yRange=[0, 1])
def disable_mouse(box_scene):
box_scene.setMouseEnabled(False, False)
box_scene.setMenuEnabled(False)
disable_mouse(self.box_scene)
disable_mouse(self.plot_prob)
self.tooltip_items = []
pen = QPen(self.palette().color(QPalette.Text))
for axis in ("left", "bottom"):
self.ploti.getAxis(axis).setPen(pen)
self._legend = LegendItem()
self._legend.setParentItem(self.box_scene)
self._legend.hide()
self._legend.anchor((1, 0), (1, 0))
self.test_changed()
def update_views(self):
"""
resize
"""
self.plot_prob.setGeometry(self.box_scene.sceneBoundingRect())
self.plot_prob.linkedViewChanged(self.box_scene, self.plot_prob.XAxis)
def set_data(self, data):
self.closeContext()
self.clear()
self.warning()
self.data = data
self.own_distribution_choose.hide()
if data is not None:
self.available_columns[:] = data.domain
domain = self.data.domain
self.varmodel[:] = list(domain) + [
meta for meta in domain.metas
if meta.is_continuous or meta.is_discrete
]
self.groupvarmodel = \
["(None)"] + [var for var in domain if var.is_discrete] + \
[meta for meta in domain.metas if meta.is_discrete]
if domain.has_discrete_class:
self.groupvar_idx = \
self.groupvarmodel[1:].index(domain.class_var) + 1
self.openContext(domain)
self.column_idx = min(max(self.column_idx, 0),
len(self.varmodel) - 1)
self.groupvar_idx = min(max(self.groupvar_idx, 0),
len(self.groupvarmodel) - 1)
self._setup()
def test_changed(self):
"""
Management of buttons, depends from type of distribution test.
Own samples are hidden.
:return:
"""
for idx, button in enumerate(self.distribution_choose.buttons):
if Distribution(button.text()) in self.test.allowed_distribution:
button.show()
if self.distribution not in self.test.allowed_distribution:
button.toggle()
self.distribution_idx = idx
else:
self.own_distribution_choose.hide()
button.hide()
self.compute_p_value()
def distribution_changed(self):
"""
Control buttons of allowed distributions - show or hide
:return: compute p-value
"""
if self.distribution == Distribution.OWN:
self.own_distribution_choose.show()
else:
self.own_distribution_choose.hide()
self.compute_p_value()
def clear(self):
self.box_scene.clear()
self.plot_prob.clear()
self.varmodel[:] = []
self.groupvarmodel = []
self.column_idx = -1
self.groupvar_idx = 0
self._legend.clear()
self._legend.hide()
def column_changed(self):
"""
compute p-value if column is changed
"""
self._setup()
self.compute_p_value()
def _setup(self):
"""
set new plot
"""
self.box_scene.clear()
self.plot_prob.clear()
self._legend.clear()
self._legend.hide()
varidx = self.column_idx
self.var = self.cvar = None
if varidx >= 0:
self.var = self.varmodel[varidx]
data = self.data
if self.var is None:
return
self.set_left_axis_name()
if self.cvar:
self.contingencies = \
contingency.get_contingency(data, self.var, self.cvar)
self.display_contingency()
else:
self.distributions = \
distribution.get_distribution(data, self.var)
self.display_distribution()
self.box_scene.autoRange()
def compute_p_value(self):
"""
:return: p-value
"""
if self.data is not None:
p_value = self.test.compute(self)
if isinstance(p_value, float):
self.infolabel.setText('\np-value: {}'.format(p_value))
else:
self.infolabel.setText('\np-value: {}'.format(round(
p_value, 3)))
def _on_relative_freq_changed(self):
self.set_left_axis_name()
if self.cvar and self.cvar.is_discrete:
self.display_contingency()
else:
self.display_distribution()
self.box_scene.autoRange()
def display_contingency(self):
"""
Set the contingency to display.
"""
cont = self.contingencies
var, cvar = self.var, self.cvar
assert len(cont) > 0
self.box_scene.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(np.sum(W))
cols.append(colors[i])
cvar_values.append(cvar.values[i])
curves.append(
ash_curve(dist, cont, m=DistributionTest.ASH_HIST))
weights = np.array(weights)
sumw = np.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 += np.array(((X[1] - X[0]) / 2)[:-1])
Y = np.array(Y)
curvesline.append((X, Y))
for t in ["fill", "line"]:
for (X, Y), color, w, cval in reversed(
list(zip(curvesline, colors, weights, cvar_values))):
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.box_scene.addItem(item)
if t == "line":
if self.relative_freq:
density = "Normalized density"
else:
density = "Density"
item.tooltip = "{density}\n{name}={value}".format(
value=cval, name=cvar.name, density=density)
self.tooltip_items.append((self.box_scene, item))
if self.show_prob:
all_X = np.array(
np.unique(np.hstack([X for X, _ in curvesline])))
inter_X = np.array(
np.linspace(all_X[0], all_X[-1],
len(all_X) * 2))
curvesinterp = [
np.interp(inter_X, X, Y) for (X, Y) in curvesline
]
sumprob = np.sum(curvesinterp, axis=0)
legal = sumprob > 0.05 * np.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{name}={value}".format(
name=cvar.name, value=cval)
self.tooltip_items.append((self.plot_prob, item))
elif var and var.is_discrete:
bottomaxis.setTicks([list(enumerate(var.values))])
cont = np.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)
item = DistributionBarItem(
geom, dist / scvar /
maxrh if self.relative_freq else dist / maxh, colors)
self.box_scene.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))
if self.relative_freq:
frequency = "Normalized frequency"
else:
frequency = "Frequency"
item.tooltip = \
"{frequency}({name}={value}):\n{tooltip}".format(
frequency=frequency,
name=cvar.name,
value=value,
tooltip=tooltip,
)
self.tooltip_items.append((self.box_scene, 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)) \
or dsum < 1e-6:
continue
position = -0.333 + ((ic + 0.5) * 0.666 / len(dist))
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()
bar = pg.ErrorBarItem()
pen = QPen(QBrush(QColor(0)), 1)
pen.setCosmetic(True)
bar.setData(x=[i + position],
y=[prob],
bottom=min(np.array([ci]), prob),
top=min(np.array([ci]), 1 - prob),
beam=np.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(bar)
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()
def set_left_axis_name(self):
leftaxis = self.ploti.getAxis("left")
set_label = leftaxis.setLabel
if self.var and self.var.is_continuous:
set_label(["Density", "Relative density"][self.cvar is not None
and self.relative_freq])
else:
set_label(["Frequency",
"Relative frequency"][self.cvar is not None
and self.relative_freq])
leftaxis.resizeEvent()
def display_distribution(self):
dist = self.distributions
var = self.var
assert len(dist) > 0
self.box_scene.clear()
self.plot_prob.clear()
self.ploti.hideAxis('right')
self.tooltip_items = []
bottomaxis = self.ploti.getAxis("bottom")
bottomaxis.setLabel(var.name)
bottomaxis.resizeEvent()
self.set_left_axis_name()
if var and var.is_continuous:
bottomaxis.setTicks(None)
if not len(dist[0]):
return
edges, curve = ash_curve(dist, None, m=DistributionTest.ASH_HIST)
edges = edges + (edges[1] - edges[0]) / 2
edges = edges[:-1]
item = pg.PlotCurveItem()
pen = QPen(QBrush(Qt.black), 3)
pen.setCosmetic(True)
item.setData(edges,
curve,
antialias=True,
stepMode=False,
fillLevel=0,
brush=QBrush(Qt.gray),
pen=pen)
self.box_scene.addItem(item)
item.tooltip = "Density"
self.tooltip_items.append((self.box_scene, item))
else:
bottomaxis.setTicks([list(enumerate(var.values))])
for i, w in enumerate(dist):
geom = QRectF(i - 0.33, 0, 0.66, w)
item = DistributionBarItem(geom, [1.0],
[QColor(128, 128, 128)])
self.box_scene.addItem(item)
item.tooltip = "Frequency for %s: %r" % (var.values[i], w)
self.tooltip_items.append((self.box_scene, item))
def onDeleteWidget(self):
self.box_scene.clear()
super().onDeleteWidget()
def get_widget_name_extension(self):
if self.column_idx >= 0:
return self.varmodel[self.column_idx]
@property
def test(self) -> Test:
return self.available_tests[self.test_idx]
@property
def distribution(self) -> Distribution:
return self.available_distributions[self.distribution_idx]
@property
def column(self):
return self.data[:, self.column_idx]
@property
def own_distribution(self):
return self.data[:, self.own_distribution_idx]
def __init__(self, parent=None):
super().__init__(parent)
self.data = None
self.distributions = None
self.contingencies = None
self.var = self.cvar = None
varbox = gui.widgetBox(self.controlArea, "Variable")
self.varmodel = itemmodels.VariableListModel()
self.groupvarmodel = itemmodels.VariableListModel()
self.varview = QtGui.QListView(
selectionMode=QtGui.QListView.SingleSelection)
self.varview.setSizePolicy(
QtGui.QSizePolicy.Minimum, QtGui.QSizePolicy.Expanding)
self.varview.setModel(self.varmodel)
self.varview.setSelectionModel(
itemmodels.ListSingleSelectionModel(self.varmodel))
self.varview.selectionModel().selectionChanged.connect(
self._on_variable_idx_changed)
varbox.layout().addWidget(self.varview)
gui.separator(varbox, 8, 8)
gui.comboBox(
varbox, self, "cont_est_type", label="Show continuous variables by",
valueType=int,
items=["Histograms", "Average shifted histograms",
"Kernel density estimators"],
callback=self._on_cont_est_type_changed)
box = gui.widgetBox(self.controlArea, "Group by")
self.groupvarview = QtGui.QListView(
selectionMode=QtGui.QListView.SingleSelection)
self.groupvarview.setFixedHeight(100)
self.groupvarview.setSizePolicy(
QtGui.QSizePolicy.Minimum, QtGui.QSizePolicy.Preferred)
self.groupvarview.setModel(self.groupvarmodel)
self.groupvarview.selectionModel().selectionChanged.connect(
self._on_groupvar_idx_changed)
box.layout().addWidget(self.groupvarview)
self.cb_rel_freq = gui.checkBox(
box, self, "relative_freq", "Show relative frequencies",
callback=self._on_relative_freq_changed)
plotview = pg.PlotWidget(background=None)
self.mainArea.layout().addWidget(plotview)
w = QtGui.QLabel()
w.setSizePolicy(QtGui.QSizePolicy.Expanding, QtGui.QSizePolicy.Fixed)
self.mainArea.layout().addWidget(w, Qt.AlignCenter)
self.plot = pg.PlotItem()
# self.plot.getViewBox().setMouseEnabled(False, False)
self.plot.getViewBox().setMenuEnabled(False)
plotview.setCentralItem(self.plot)
pen = QtGui.QPen(self.palette().color(QtGui.QPalette.Text))
for axis in ("left", "bottom"):
self.plot.getAxis(axis).setPen(pen)
self._legend = LegendItem()
self._legend.setParentItem(self.plot.getViewBox())
self._legend.hide()
self._legend.anchor((1, 0), (1, 0))
class OWDistributions(widget.OWWidget):
name = "Distributions"
description = "Display value distributions of a data feature in a graph."
icon = "icons/Distribution.svg"
priority = 100
inputs = [InputSignal("Data", Orange.data.Table, "set_data",
doc="Set the input data set")]
settingsHandler = settings.DomainContextHandler()
#: Selected variable index
variable_idx = settings.ContextSetting(-1)
#: Selected group variable
groupvar_idx = settings.ContextSetting(0)
Hist, ASH, Kernel = 0, 1, 2
#: Continuous variable density estimation method
cont_est_type = settings.Setting(ASH)
relative_freq = settings.Setting(False)
def __init__(self, parent=None):
super().__init__(parent)
self.data = None
self.distributions = None
self.contingencies = None
self.var = self.cvar = None
varbox = gui.widgetBox(self.controlArea, "Variable")
self.varmodel = itemmodels.VariableListModel()
self.groupvarmodel = itemmodels.VariableListModel()
self.varview = QtGui.QListView(
selectionMode=QtGui.QListView.SingleSelection)
self.varview.setSizePolicy(
QtGui.QSizePolicy.Minimum, QtGui.QSizePolicy.Expanding)
self.varview.setModel(self.varmodel)
self.varview.setSelectionModel(
itemmodels.ListSingleSelectionModel(self.varmodel))
self.varview.selectionModel().selectionChanged.connect(
self._on_variable_idx_changed)
varbox.layout().addWidget(self.varview)
gui.separator(varbox, 8, 8)
gui.comboBox(
varbox, self, "cont_est_type", label="Show continuous variables by",
valueType=int,
items=["Histograms", "Average shifted histograms",
"Kernel density estimators"],
callback=self._on_cont_est_type_changed)
box = gui.widgetBox(self.controlArea, "Group by")
self.groupvarview = QtGui.QListView(
selectionMode=QtGui.QListView.SingleSelection)
self.groupvarview.setFixedHeight(100)
self.groupvarview.setSizePolicy(
QtGui.QSizePolicy.Minimum, QtGui.QSizePolicy.Preferred)
self.groupvarview.setModel(self.groupvarmodel)
self.groupvarview.selectionModel().selectionChanged.connect(
self._on_groupvar_idx_changed)
box.layout().addWidget(self.groupvarview)
self.cb_rel_freq = gui.checkBox(
box, self, "relative_freq", "Show relative frequencies",
callback=self._on_relative_freq_changed)
plotview = pg.PlotWidget(background=None)
self.mainArea.layout().addWidget(plotview)
w = QtGui.QLabel()
w.setSizePolicy(QtGui.QSizePolicy.Expanding, QtGui.QSizePolicy.Fixed)
self.mainArea.layout().addWidget(w, Qt.AlignCenter)
self.plot = pg.PlotItem()
# self.plot.getViewBox().setMouseEnabled(False, False)
self.plot.getViewBox().setMenuEnabled(False)
plotview.setCentralItem(self.plot)
pen = QtGui.QPen(self.palette().color(QtGui.QPalette.Text))
for axis in ("left", "bottom"):
self.plot.getAxis(axis).setPen(pen)
self._legend = LegendItem()
self._legend.setParentItem(self.plot.getViewBox())
self._legend.hide()
self._legend.anchor((1, 0), (1, 0))
def set_data(self, data):
self.closeContext()
self.clear()
self.data = data
if self.data is not None:
domain = self.data.domain
self.varmodel[:] = list(domain)
self.groupvarmodel[:] = \
["(None)"] + [var for var in domain if var.is_discrete]
if domain.has_discrete_class:
self.groupvar_idx = \
list(self.groupvarmodel).index(domain.class_var)
self.openContext(domain)
self.variable_idx = min(max(self.variable_idx, 0),
len(self.varmodel) - 1)
self.groupvar_idx = min(max(self.groupvar_idx, 0),
len(self.groupvarmodel) - 1)
itemmodels.select_row(self.groupvarview, self.groupvar_idx)
itemmodels.select_row(self.varview, self.variable_idx)
self._setup()
def clear(self):
self.plot.clear()
self.varmodel[:] = []
self.groupvarmodel[:] = []
self.variable_idx = -1
self.groupvar_idx = 0
self._legend.clear()
self._legend.hide()
def _setup(self):
self.plot.clear()
self._legend.clear()
self._legend.hide()
varidx = self.variable_idx
self.var = self.cvar = None
if varidx >= 0:
self.var = self.varmodel[varidx]
if self.groupvar_idx > 0:
self.cvar = self.groupvarmodel[self.groupvar_idx]
self.set_left_axis_name()
self.enable_disable_rel_freq()
if self.var is None:
return
if self.cvar:
self.contingencies = \
contingency.get_contingency(self.data, self.var, self.cvar)
self.display_contingency()
else:
self.distributions = \
distribution.get_distribution(self.data, self.var)
self.display_distribution()
def _density_estimator(self):
if self.cont_est_type == OWDistributions.Hist:
def hist(dist):
h, edges = numpy.histogram(dist[0, :], bins=10,
weights=dist[1, :])
return edges, h
return hist
elif self.cont_est_type == OWDistributions.ASH:
return lambda dist: ash_curve(dist, m=5)
elif self.cont_est_type == OWDistributions.Kernel:
return rect_kernel_curve
def display_distribution(self):
dist = self.distributions
var = self.var
assert len(dist) > 0
self.plot.clear()
bottomaxis = self.plot.getAxis("bottom")
bottomaxis.setLabel(var.name)
self.set_left_axis_name()
if var and var.is_continuous:
bottomaxis.setTicks(None)
curve_est = self._density_estimator()
edges, curve = curve_est(dist)
item = pg.PlotCurveItem()
item.setData(edges, curve, antialias=True, stepMode=True,
fillLevel=0, brush=QtGui.QBrush(Qt.gray),
pen=QtGui.QColor(Qt.white))
self.plot.addItem(item)
else:
bottomaxis.setTicks([list(enumerate(var.values))])
for i, w in enumerate(dist):
geom = QtCore.QRectF(i - 0.33, 0, 0.66, w)
item = DistributionBarItem(geom, [1.0],
[QtGui.QColor(128, 128, 128)])
self.plot.addItem(item)
def _on_relative_freq_changed(self):
self.set_left_axis_name()
if self.cvar and self.cvar.is_discrete:
self.display_contingency()
else:
self.display_distribution()
def display_contingency(self):
"""
Set the contingency to display.
"""
cont = self.contingencies
var, cvar = self.var, self.cvar
assert len(cont) > 0
self.plot.clear()
self._legend.clear()
bottomaxis = self.plot.getAxis("bottom")
bottomaxis.setLabel(var.name)
palette = colorpalette.ColorPaletteGenerator(len(cvar.values))
colors = [palette[i] for i in range(len(cvar.values))]
if var and var.is_continuous:
bottomaxis.setTicks(None)
weights = numpy.array([numpy.sum(W) for _, W in cont])
weights /= numpy.sum(weights)
curve_est = self._density_estimator()
curves = [curve_est(dist) for dist in cont]
curves = [(X, Y * w) for (X, Y), w in zip(curves, weights)]
cum_curves = [curves[0]]
for X, Y in curves[1:]:
cum_curves.append(sum_rect_curve(X, Y, *cum_curves[-1]))
for (X, Y), color in reversed(list(zip(cum_curves, colors))):
item = pg.PlotCurveItem()
pen = QtGui.QPen(QtGui.QBrush(Qt.white), 0.5)
pen.setCosmetic(True)
item.setData(X, Y, antialias=True, stepMode=True,
fillLevel=0, brush=QtGui.QBrush(color.lighter()),
pen=pen)
self.plot.addItem(item)
# # XXX: sum the individual curves and not the distributions.
# # The conditional distributions might be 'smoother' than
# # the cumulative one
# cum_dist = [cont[0]]
# for dist in cont[1:]:
# cum_dist.append(dist_sum(dist, cum_dist[-1]))
#
# curves = [rect_kernel_curve(dist) for dist in cum_dist]
# colors = [Qt.blue, Qt.red, Qt.magenta]
# for (X, Y), color in reversed(list(zip(curves, colors))):
# item = pg.PlotCurveItem()
# item.setData(X, Y, antialias=True, stepMode=True,
# fillLevel=0, brush=QtGui.QBrush(color))
# item.setPen(QtGui.QPen(color))
# self.plot.addItem(item)
elif var and var.is_discrete:
bottomaxis.setTicks([list(enumerate(var.values))])
cont = numpy.array(cont)
for i, (value, dist) in enumerate(zip(var.values, cont.T)):
dsum = sum(dist)
geom = QtCore.QRectF(i - 0.333, 0, 0.666, 100
if self.relative_freq else dsum)
item = DistributionBarItem(geom, dist / dsum, colors)
self.plot.addItem(item)
for color, name in zip(colors, cvar.values):
self._legend.addItem(
ScatterPlotItem(pen=color, brush=color, size=10, shape="s"),
name
)
self._legend.show()
def set_left_axis_name(self):
set_label = self.plot.getAxis("left").setLabel
if (self.var and
self.var.is_continuous and
self.cont_est_type != OWDistributions.Hist):
set_label("Density")
else:
set_label(["Frequency", "Relative frequency"]
[self.cvar is not None and self.relative_freq])
def enable_disable_rel_freq(self):
self.cb_rel_freq.setDisabled(
self.var is None or self.cvar is None or self.var.is_continuous)
def _on_variable_idx_changed(self):
self.variable_idx = selected_index(self.varview)
self._setup()
def _on_groupvar_idx_changed(self):
self.groupvar_idx = selected_index(self.groupvarview)
self._setup()
def _on_cont_est_type_changed(self):
self.set_left_axis_name()
if self.data is not None:
self._setup()
def onDeleteWidget(self):
self.plot.clear()
super().onDeleteWidget()
