class RadioBooleanFilter(QWidget, Control):
""" Boolean filter (Only/Exclude)
"""
def __init__(self, tree, dataset, master, parent=None):
QWidget.__init__(self, parent)
Control.__init__(self, tree, dataset, master)
self.setLayout(QVBoxLayout())
self.buttonGroup = QButtonGroup(self)
self.values = []
for i, option in enumerate(tree.subelements_top("Option")):
rb = QRadioButton(option.displayName, self)
self.buttonGroup.addButton(rb)
self.buttonGroup.setId(rb, i)
self.layout().addWidget(rb)
self.values.append(option.value)
self.buttonGroup.button(0).setChecked(True)
def value(self):
return {"excluded": "%i" % self.buttonGroup.checkedId()}
def get_filter(self):
return self.tree.internalName, self.value()
def query(self):
return [("Filter", self.tree, self.value())]
def setControlValue(self, name, value):
for i, v in enumerate(self.values):
if v == value:
button = self.buttonGroup.button(i)
button.setChecked(True)
break
def __init__(self):
super().__init__()
self.corpus = None
self.learning_thread = None
# Commit button
gui.auto_commit(self.buttonsArea, self, 'autocommit', 'Commit', box=False)
button_group = QButtonGroup(self, exclusive=True)
button_group.buttonClicked[int].connect(self.change_method)
self.widgets = []
method_layout = QVBoxLayout()
self.controlArea.layout().addLayout(method_layout)
for i, (method, attr_name) in enumerate(self.methods):
widget = method(self, title='Options')
widget.setFixedWidth(self.control_area_width)
widget.valueChanged.connect(self.commit)
self.widgets.append(widget)
setattr(self, attr_name, widget)
rb = QRadioButton(text=widget.Model.name)
button_group.addButton(rb, i)
method_layout.addWidget(rb)
method_layout.addWidget(widget)
button_group.button(self.method_index).setChecked(True)
self.toggle_widgets()
method_layout.addStretch()
# Topics description
self.topic_desc = TopicViewer()
self.topic_desc.topicSelected.connect(self.send_topic_by_id)
self.mainArea.layout().addWidget(self.topic_desc)
self.topic_desc.setFocus()
def __init__(self):
super().__init__()
self.corpus = None
self.learning_thread = None
# Commit button
gui.auto_commit(self.buttonsArea, self, 'autocommit', 'Commit', box=False)
button_group = QButtonGroup(self, exclusive=True)
button_group.buttonClicked[int].connect(self.change_method)
self.widgets = []
method_layout = QVBoxLayout()
self.controlArea.layout().addLayout(method_layout)
for i, (method, attr_name) in enumerate(self.methods):
widget = method(self, title='Options')
widget.setFixedWidth(self.control_area_width)
widget.valueChanged.connect(self.commit)
self.widgets.append(widget)
setattr(self, attr_name, widget)
rb = QRadioButton(text=widget.Model.name)
button_group.addButton(rb, i)
method_layout.addWidget(rb)
method_layout.addWidget(widget)
button_group.button(self.method_index).setChecked(True)
self.toggle_widgets()
method_layout.addStretch()
# Topics description
self.topic_desc = TopicViewer()
self.topic_desc.topicSelected.connect(self.send_topic_by_id)
self.mainArea.layout().addWidget(self.topic_desc)
self.topic_desc.setFocus()
class RadioBooleanFilter(QWidget, Control):
""" Boolean filter (Only/Exclude)
"""
def __init__(self, tree, dataset, master, parent=None):
QWidget.__init__(self, parent)
Control.__init__(self, tree, dataset, master)
self.setLayout(QVBoxLayout())
self.buttonGroup = QButtonGroup(self)
self.values = []
for i, option in enumerate(tree.subelements_top("Option")):
rb = QRadioButton(option.displayName, self)
self.buttonGroup.addButton(rb)
self.buttonGroup.setId(rb, i)
self.layout().addWidget(rb)
self.values.append(option.value)
self.buttonGroup.button(0).setChecked(True)
def value(self):
return {"excluded": "%i" % self.buttonGroup.checkedId()}
def get_filter(self):
return self.tree.internalName, self.value()
def query(self):
return [("Filter", self.tree, self.value())]
def setControlValue(self, name, value):
for i, v in enumerate(self.values):
if v == value:
button = self.buttonGroup.button(i)
button.setChecked(True)
break
def __init__(self):
super().__init__()
ConcurrentWidgetMixin.__init__(self)
self.corpus = None
self.learning_thread = None
self.__pending_selection = self.selection
self.perplexity = "n/a"
self.coherence = "n/a"
# Commit button
gui.auto_commit(self.buttonsArea,
self,
'autocommit',
'Commit',
box=False)
button_group = QButtonGroup(self, exclusive=True)
button_group.buttonClicked[int].connect(self.change_method)
self.widgets = []
method_layout = QVBoxLayout()
self.controlArea.layout().addLayout(method_layout)
for i, (method, attr_name) in enumerate(self.methods):
widget = method(self, title='Options')
widget.setFixedWidth(self.control_area_width)
widget.valueChanged.connect(self.commit.deferred)
self.widgets.append(widget)
setattr(self, attr_name, widget)
rb = QRadioButton(text=widget.Model.name)
button_group.addButton(rb, i)
method_layout.addWidget(rb)
method_layout.addWidget(widget)
button_group.button(self.method_index).setChecked(True)
self.toggle_widgets()
method_layout.addStretch()
box = gui.vBox(self.controlArea, "Topic evaluation")
gui.label(box, self, "Log perplexity: %(perplexity)s")
gui.label(box, self, "Topic coherence: %(coherence)s")
self.controlArea.layout().insertWidget(1, box)
# Topics description
self.topic_desc = TopicViewer()
self.topic_desc.topicSelected.connect(self.send_topic_by_id)
self.mainArea.layout().addWidget(self.topic_desc)
self.topic_desc.setFocus()
def qbuttongroup_emit_clicked(bg: QButtonGroup, id_: int):
button = bg.button(id_)
bg.buttonClicked.emit(button)
if QT_VERSION_INFO >= (5, 15):
bg.idClicked.emit(id_)
if QT_VERSION_INFO < (6, 0):
bg.buttonClicked[int].emit(id_)
class SingleMethodModule(PreprocessorModule):
Methods = NotImplemented
DEFAULT_METHOD = NotImplemented
def __init__(self, parent=None, **kwargs):
super().__init__(parent, **kwargs)
self.__method = self.DEFAULT_METHOD
self.setLayout(QGridLayout())
self.__group = QButtonGroup(self, exclusive=True)
self.__group.buttonClicked.connect(self.__method_rb_clicked)
for method_id in range(len(self.Methods)):
method = self.Methods[method_id]
rb = QRadioButton(method.name)
rb.setChecked(self.__method == method_id)
rb.setToolTip(self.get_tooltip(method))
self.__group.addButton(rb, method_id)
self.layout().addWidget(rb)
@property
def method(self) -> int:
return self.__method
def setParameters(self, params: Dict):
self._set_method(params.get("method", self.DEFAULT_METHOD))
def _set_method(self, method: int):
if self.__method != method:
self.__method = method
self.__group.button(method).setChecked(True)
self.changed.emit()
def __method_rb_clicked(self):
self._set_method(self.__group.checkedId())
self.edited.emit()
def parameters(self) -> Dict:
return {"method": self.__method}
def __repr__(self):
return self.Methods[self.__method].name
class OWDiscretize(widget.OWWidget):
# pylint: disable=too-many-instance-attributes
name = "Discretize"
description = "Discretize numeric variables"
category = "Transform"
icon = "icons/Discretize.svg"
keywords = ["bin", "categorical", "nominal", "ordinal"]
priority = 2130
class Inputs:
data = Input("Data", Table, doc="Input data table")
class Outputs:
data = Output("Data", Table, doc="Table with categorical features")
settings_version = 3
#: Default setting (key DefaultKey) and specific settings for variables;
# if variable is not in the dict, it uses default
var_hints: Dict[KeyType, VarHint] = Setting(
{DefaultKey: DefaultHint}, schema_only=True)
autosend = Setting(True)
want_main_area = False
def __init__(self):
super().__init__()
#: input data
self.data = None
#: Cached discretized variables
self.discretized_vars: Dict[KeyType, DiscreteVariable] = {}
# Indicates that buttons, spins, edit and combos are being changed
# programmatically (when interface is changed due to selection change),
# so this should not trigger update of hints and invalidation of
# discretization in `self.discretized_vars`.
self.__interface_update = False
box = gui.hBox(self.controlArea, True, spacing=8)
self._create_var_list(box)
self._create_buttons(box)
gui.auto_apply(self.buttonsArea, self, "autosend")
gui.rubber(self.buttonsArea)
self.varview.select_default()
def _create_var_list(self, box):
"""Create list view with variables"""
# If we decide to not elide, remove the `uniformItemSize` argument
self.varview = ListViewSearch(
selectionMode=QListView.ExtendedSelection, uniformItemSizes=True)
self.varview.setModel(
DiscDomainModel(
valid_types=(ContinuousVariable, TimeVariable),
order=DiscDomainModel.MIXED
))
self.varview.selectionModel().selectionChanged.connect(
self._var_selection_changed)
self.varview.default_view.selectionModel().selectionChanged.connect(
self._default_selected)
self._update_default_model()
box.layout().addWidget(self.varview)
def _create_buttons(self, box):
"""Create radio buttons"""
def intspin():
s = QSpinBox(self)
s.setMinimum(2)
s.setMaximum(10)
s.setFixedWidth(60)
s.setAlignment(Qt.AlignRight)
s.setContentsMargins(0, 0, 0, 0)
return s, s.valueChanged
def widthline(validator):
s = QLineEdit(self)
s.setFixedWidth(60)
s.setAlignment(Qt.AlignRight)
s.setValidator(validator)
s.setContentsMargins(0, 0, 0, 0)
return s, s.textChanged
def manual_cut_editline(text="", enabled=True) -> QLineEdit:
edit = QLineEdit(
text=text,
placeholderText="e.g. 0.0, 0.5, 1.0",
toolTip='<p style="white-space:pre">' +
'Enter cut points as a comma-separate list of \n'
'strictly increasing numbers e.g. 0.0, 0.5, 1.0).</p>',
enabled=enabled,
)
edit.setValidator(IncreasingNumbersListValidator())
edit.setSizePolicy(QSizePolicy.MinimumExpanding, QSizePolicy.Fixed)
@edit.textChanged.connect
def update():
validator = edit.validator()
if validator is not None and edit.text().strip():
state, _, _ = validator.validate(edit.text(), 0)
else:
state = QValidator.Acceptable
palette = edit.palette()
colors = {
QValidator.Intermediate: (Qt.yellow, Qt.black),
QValidator.Invalid: (Qt.red, Qt.black),
}.get(state, None)
if colors is None:
palette = QPalette()
else:
palette.setColor(QPalette.Base, colors[0])
palette.setColor(QPalette.Text, colors[1])
cr = edit.cursorRect()
p = edit.mapToGlobal(cr.bottomRight())
edit.setPalette(palette)
if state != QValidator.Acceptable and edit.isVisible():
validator.show_tip(edit, p, edit.toolTip(),
textFormat=Qt.RichText)
else:
validator.show_tip(edit, p, "")
return edit, edit.textChanged
children = []
def button(id_, *controls, stretch=True):
layout = QHBoxLayout()
desc = Options[id_]
button = QRadioButton(desc.label)
button.setToolTip(desc.tooltip)
self.button_group.addButton(button, id_)
layout.addWidget(button)
if controls:
if stretch:
layout.addStretch(1)
for c, signal in controls:
layout.addWidget(c)
if signal is not None:
@signal.connect
def arg_changed():
self.button_group.button(id_).setChecked(True)
self.update_hints(id_)
children.append(layout)
button_box.layout().addLayout(layout)
return (*controls, (None, ))[0][0]
button_box = gui.vBox(box)
button_box.layout().setSpacing(0)
button_box.setSizePolicy(QSizePolicy(QSizePolicy.Fixed, QSizePolicy.Preferred))
self.button_group = QButtonGroup(self)
self.button_group.idClicked.connect(self.update_hints)
button(Methods.Keep)
button(Methods.Remove)
self.binning_spin = button(Methods.Binning, intspin())
validator = QDoubleValidator()
validator.setBottom(0)
self.width_line = button(Methods.FixedWidth, widthline(validator))
self.width_time_unit = u = QComboBox(self)
u.setContentsMargins(0, 0, 0, 0)
u.addItems([unit + "(s)" for unit in time_units])
validator = QIntValidator()
validator.setBottom(1)
self.width_time_line = button(Methods.FixedWidthTime,
widthline(validator),
(u, u.currentTextChanged))
self.freq_spin = button(Methods.EqualFreq, intspin())
self.width_spin = button(Methods.EqualWidth, intspin())
button(Methods.MDL)
self.copy_to_custom = FixedSizeButton(
text="CC", toolTip="Copy the current cut points to manual mode")
self.copy_to_custom.clicked.connect(self._copy_to_manual)
self.threshold_line = button(Methods.Custom,
manual_cut_editline(),
(self.copy_to_custom, None),
stretch=False)
button(Methods.Default)
maxheight = max(w.sizeHint().height() for w in children)
for w in children:
w.itemAt(0).widget().setFixedHeight(maxheight)
button_box.layout().addStretch(1)
def _update_default_model(self):
"""Update data in the model showing default settings"""
model = self.varview.default_view.model()
model.setData(model.index(0), self.var_hints[DefaultKey], Qt.UserRole)
def _set_mdl_button(self):
"""Disable MDL discretization for data with non-discrete class"""
mdl_button = self.button_group.button(Methods.MDL)
if self.data is None or self.data.domain.has_discrete_class:
mdl_button.setEnabled(True)
else:
if mdl_button.isChecked():
self._check_button(Methods.Keep, True)
mdl_button.setEnabled(False)
def _check_button(self, method_id: Methods, checked: bool):
"""Checks the given button"""
self.button_group.button(method_id).setChecked(checked)
def _uncheck_all_buttons(self):
"""Uncheck all radio buttons"""
group = self.button_group
button = group.checkedButton()
if button is not None:
group.setExclusive(False)
button.setChecked(False)
group.setExclusive(True)
def _set_radio_enabled(self, method_id: Methods, value: bool):
"""Enable/disable radio button and related controls"""
if self.button_group.button(method_id).isChecked() and not value:
self._uncheck_all_buttons()
self.button_group.button(method_id).setEnabled(value)
for control_name in Options[method_id].controls:
getattr(self, control_name).setEnabled(value)
def _get_values(self, method_id: Methods) -> Tuple[Union[int, float, str]]:
"""Return parameters from controls pertaining to the given method"""
controls = Options[method_id].controls
values = []
for control_name in controls:
control = getattr(self, control_name)
if isinstance(control, QSpinBox):
values.append(control.value())
elif isinstance(control, QComboBox):
values.append(control.currentIndex())
else:
values.append(control.text())
return tuple(values)
def _set_values(self, method_id: Methods,
values: Tuple[Union[str, int, float]]):
"""
Set controls pertaining to the given method to parameters from hint
"""
controls = Options[method_id].controls
for control_name, value in zip(controls, values):
control = getattr(self, control_name)
if isinstance(control, QSpinBox):
control.setValue(value)
elif isinstance(control, QComboBox):
control.setCurrentIndex(value)
else:
control.setText(value)
def varkeys_for_selection(self) -> List[KeyType]:
"""
Return list of KeyType's for selected variables (for indexing var_hints)
If 'Default settings' are selected, this returns DefaultKey
"""
model = self.varview.model()
varkeys = [variable_key(model[index.row()])
for index in self.varview.selectionModel().selectedRows()]
return varkeys or [DefaultKey] # default settings are selected
def update_hints(self, method_id: Methods):
"""
Callback for radio buttons and for controls regulating parameters
This function:
- updates `var_hints` for all selected methods
- invalidates (removes) `discretized_vars` for affected variables
- calls _update_discretizations to compute and commit new discretization
- calls deferred commit
Data for list view models is updated in _update_discretizations
"""
if self.__interface_update:
return
method_id = Methods(method_id)
args = self._get_values(method_id)
keys = self.varkeys_for_selection()
if method_id == Methods.Default:
for key in keys:
if key in self.var_hints:
del self.var_hints[key]
else:
self.var_hints.update(dict.fromkeys(keys, VarHint(method_id, args)))
if keys == [DefaultKey]:
invalidate = set(self.discretized_vars) - set(self.var_hints)
else:
invalidate = keys
for key in invalidate:
del self.discretized_vars[key]
if keys == [DefaultKey]:
self._update_default_model()
self._update_discretizations()
self.commit.deferred()
def _update_discretizations(self):
"""
Compute invalidated (missing) discretizations
Also set data for list view models for all invalidated variables
"""
if self.data is None:
return
default_hint = self.var_hints[DefaultKey]
model = self.varview.model()
for index, var in enumerate(model):
key = variable_key(var)
if key in self.discretized_vars:
continue # still valid
var_hint = self.var_hints.get(key)
points, dvar = self._discretize_var(var, var_hint or default_hint)
self.discretized_vars[key] = dvar
values = getattr(dvar, "values", ())
model.setData(model.index(index),
DiscDesc(var_hint, points, values),
Qt.UserRole)
def _discretize_var(self, var: ContinuousVariable, hint: VarHint) \
-> Tuple[str, Optional[Variable]]:
"""
Discretize using method and data in the hint.
Returns a description (list of points or error/warning) and a
- discrete variable
- same variable (if kept numeric)
- None (if removed or errored)
"""
if isinstance(var, TimeVariable):
if hint.method_id in (Methods.FixedWidth, Methods.Custom):
return ": <keep, time var>", var
else:
if hint.method_id == Methods.FixedWidthTime:
return ": <keep, not time>", var
function = Options[hint.method_id].function
dvar = function(self.data, var, *hint.args)
if isinstance(dvar, str):
return f" <{dvar}>", None # error
if dvar is None:
return "", None # removed
elif dvar is var:
return "", var # no transformation
thresholds = dvar.compute_value.points
if len(thresholds) == 0:
return " <removed>", None
return ": " + ", ".join(map(var.repr_val, thresholds)), dvar
def _copy_to_manual(self):
"""
Callback for 'CC' button
Sets selected variables' method to "Custom" and copies thresholds
to their VarHints. Variables that are not discretized (for any reason)
are skipped.
Discretizations are invalidated and then updated
(`_update_discretizations`).
If all selected variables have the same thresholds, it copies it to
the line edit. Otherwise it unchecks all radio buttons to keep the
interface consistent.
"""
varkeys = self.varkeys_for_selection()
texts = set()
for key in varkeys:
dvar = self.discretized_vars.get(key)
fmt = self.data.domain[key[0]].repr_val
if isinstance(dvar, DiscreteVariable):
text = ", ".join(map(fmt, dvar.compute_value.points))
texts.add(text)
self.var_hints[key] = VarHint(Methods.Custom, (text, ))
del self.discretized_vars[key]
try:
self.__interface_update = True
if len(texts) == 1:
self.threshold_line.setText(texts.pop())
else:
self._uncheck_all_buttons()
finally:
self.__interface_update = False
self._update_discretizations()
self.commit.deferred()
def _default_selected(self, selected):
"""Callback for selecting 'Default setting'"""
if not selected:
# Prevent infinite recursion (with _var_selection_changed)
return
self.varview.selectionModel().clearSelection()
self._update_interface()
set_enabled = self._set_radio_enabled
set_enabled(Methods.Default, False)
set_enabled(Methods.FixedWidth, True)
set_enabled(Methods.FixedWidthTime, True)
set_enabled(Methods.Custom, True)
self.copy_to_custom.setEnabled(False)
def _var_selection_changed(self, _):
"""Callback for changed selection in listview with variables"""
selected = self.varview.selectionModel().selectedIndexes()
if not selected:
# Prevent infinite recursion (with _default_selected)
return
self.varview.default_view.selectionModel().clearSelection()
self._update_interface()
set_enabled = self._set_radio_enabled
vars_ = [self.data.domain[name]
for name, _ in self.varkeys_for_selection()]
no_time = not any(isinstance(var, TimeVariable) for var in vars_)
all_time = all(isinstance(var, TimeVariable) for var in vars_)
set_enabled(Methods.Default, True)
set_enabled(Methods.FixedWidth, no_time)
set_enabled(Methods.Custom, no_time)
self.copy_to_custom.setEnabled(no_time)
set_enabled(Methods.FixedWidthTime, all_time)
def _update_interface(self):
"""
Update the user interface according to selection
- If VarHints for all selected variables are the same, check the
corresponding radio button and fill the corresponding controls;
- otherwise, uncheck all radios.
"""
if self.__interface_update:
return
try:
self.__interface_update = True
keys = self.varkeys_for_selection()
mset = list(unique_everseen(map(self.var_hints.get, keys)))
if len(mset) != 1:
self._uncheck_all_buttons()
return
if mset == [None]:
method_id, args = Methods.Default, ()
else:
method_id, args = mset.pop()
self._check_button(method_id, True)
self._set_values(method_id, args)
finally:
self.__interface_update = False
@Inputs.data
def set_data(self, data: Optional[Table]):
self.discretized_vars = {}
self.data = data
self.varview.model().set_domain(None if data is None else data.domain)
self._update_discretizations()
self._update_default_model()
self.varview.select_default()
self._set_mdl_button()
self.commit.now()
@gui.deferred
def commit(self):
if self.data is None:
self.Outputs.data.send(None)
return
def part(variables: List[Variable]) -> List[Variable]:
return [dvar
for dvar in (self.discretized_vars.get(variable_key(v), v)
for v in variables)
if dvar]
d = self.data.domain
domain = Domain(part(d.attributes), part(d.class_vars), part(d.metas))
output = self.data.transform(domain)
self.Outputs.data.send(output)
def send_report(self):
dmodel = self.varview.default_view.model()
desc = dmodel.data(dmodel.index(0))
self.report_items((tuple(desc.split(": ", maxsplit=1)), ))
model = self.varview.model()
reported = []
for row in range(model.rowCount()):
name = model[row].name
desc = model.data(model.index(row), Qt.UserRole)
if desc.hint is not None:
name = f"{name} ({format_desc(desc.hint)})"
reported.append((name, ', '.join(desc.values)))
self.report_items("Variables", reported)
@classmethod
def migrate_settings(cls, settings, version):
if version is None or version < 2:
# was stored as int indexing Methods (but offset by 1)
default = settings.pop("default_method", 0)
default = Methods(default + 1)
settings["default_method_name"] = default.name
if version is None or version < 3:
method_name = settings.pop("default_method_name",
DefaultHint.method_id.name)
k = settings.pop("default_k", 3)
cut_points = settings.pop("default_cutpoints", ())
method_id = getattr(Methods, method_name)
if method_id in (Methods.EqualFreq, Methods.EqualWidth):
args = (k, )
elif method_id == Methods.Custom:
args = (cut_points, )
else:
args = ()
default_hint = VarHint(method_id, args)
var_hints = {DefaultKey: default_hint}
for context in settings.pop("context_settings", []):
values = context.values
if "saved_var_states" not in values:
continue
var_states, _ = values.pop("saved_var_states")
for (tpe, name), dstate in var_states.items():
key = (name, tpe == 4) # time variable == 4
method = dstate.method
method_name = type(method).__name__.replace("Leave", "Keep")
if method_name == "Default":
continue
if method_name == "Custom":
args = (", ".join(f"{x:g}" for x in method.points), )
else:
args = tuple(method)
var_hints[key] = VarHint(getattr(Methods, method_name), args)
settings["var_hints"] = var_hints
class OWRank(OWWidget):
name = "Rank"
description = "Rank and filter data features by their relevance."
icon = "icons/Rank.svg"
priority = 1102
buttons_area_orientation = Qt.Vertical
inputs = [("Data", Table, "setData"),
("Scorer", score.Scorer, "set_learner", widget.Multiple)]
outputs = [("Reduced Data", Table, widget.Default), ("Scores", Table)]
SelectNone, SelectAll, SelectManual, SelectNBest = range(4)
cls_default_selected = Setting({"Gain Ratio", "Gini Decrease"})
reg_default_selected = Setting({"Univariate Linear Regression", "RReliefF"})
selectMethod = Setting(SelectNBest)
nSelected = Setting(5)
auto_apply = Setting(True)
# Header state for discrete/continuous/no_class scores
headerState = Setting([None, None, None])
settings_version = 1
settingsHandler = DomainContextHandler()
selected_rows = ContextSetting([])
gain = inf_gain = gini = anova = chi2 = ulr = relief = rrelief = fcbc = True
_score_vars = ["gain", "inf_gain", "gini", "anova", "chi2", "relief",
"fcbc", "ulr", "rrelief"]
class Warning(OWWidget.Warning):
no_target_var = Msg("Data does not have a target variable")
class Error(OWWidget.Error):
invalid_type = Msg("Cannot handle target variable type {}")
inadequate_learner = Msg("{}")
def __init__(self):
super().__init__()
self.measure_scores = None
self.update_scores = True
self.usefulAttributes = []
self.learners = {}
self.labels = []
self.out_domain_desc = None
self.all_measures = SCORES
self.selectedMeasures = dict([(m.name, True) for m
in self.all_measures])
# Discrete (0) or continuous (1) class mode
self.rankMode = 0
self.data = None
self.discMeasures = [m for m in self.all_measures if
issubclass(DiscreteVariable, m.score.class_type)]
self.contMeasures = [m for m in self.all_measures if
issubclass(ContinuousVariable, m.score.class_type)]
self.score_checks = []
self.cls_scoring_box = gui.vBox(None, "Scoring for Classification")
self.reg_scoring_box = gui.vBox(None, "Scoring for Regression")
boxes = [self.cls_scoring_box] * 7 + [self.reg_scoring_box] * 2
for _score, var, box in zip(SCORES, self._score_vars, boxes):
check = gui.checkBox(
box, self, var, label=_score.name,
callback=lambda val=_score: self.measuresSelectionChanged(val))
self.score_checks.append(check)
self.score_stack = QStackedWidget(self)
self.score_stack.addWidget(self.cls_scoring_box)
self.score_stack.addWidget(self.reg_scoring_box)
self.score_stack.addWidget(QWidget())
self.controlArea.layout().addWidget(self.score_stack)
gui.rubber(self.controlArea)
selMethBox = gui.vBox(
self.controlArea, "Select Attributes", addSpace=True)
grid = QGridLayout()
grid.setContentsMargins(6, 0, 6, 0)
self.selectButtons = QButtonGroup()
self.selectButtons.buttonClicked[int].connect(self.setSelectMethod)
def button(text, buttonid, toolTip=None):
b = QRadioButton(text)
self.selectButtons.addButton(b, buttonid)
if toolTip is not None:
b.setToolTip(toolTip)
return b
b1 = button(self.tr("None"), OWRank.SelectNone)
b2 = button(self.tr("All"), OWRank.SelectAll)
b3 = button(self.tr("Manual"), OWRank.SelectManual)
b4 = button(self.tr("Best ranked:"), OWRank.SelectNBest)
s = gui.spin(selMethBox, self, "nSelected", 1, 100,
callback=self.nSelectedChanged)
grid.addWidget(b1, 0, 0)
grid.addWidget(b2, 1, 0)
grid.addWidget(b3, 2, 0)
grid.addWidget(b4, 3, 0)
grid.addWidget(s, 3, 1)
self.selectButtons.button(self.selectMethod).setChecked(True)
selMethBox.layout().addLayout(grid)
gui.auto_commit(selMethBox, self, "auto_apply", "Send", box=False)
# Discrete, continuous and no_class table views are stacked
self.ranksViewStack = QStackedLayout()
self.mainArea.layout().addLayout(self.ranksViewStack)
self.discRanksView = QTableView()
self.ranksViewStack.addWidget(self.discRanksView)
self.discRanksView.setSelectionBehavior(QTableView.SelectRows)
self.discRanksView.setSelectionMode(QTableView.MultiSelection)
self.discRanksView.setSortingEnabled(True)
self.discRanksLabels = ["#"] + [m.shortname for m in self.discMeasures]
self.discRanksModel = QStandardItemModel(self)
self.discRanksModel.setHorizontalHeaderLabels(self.discRanksLabels)
self.discRanksProxyModel = MySortProxyModel(self)
self.discRanksProxyModel.setSourceModel(self.discRanksModel)
self.discRanksView.setModel(self.discRanksProxyModel)
self.discRanksView.setColumnWidth(0, 20)
self.discRanksView.selectionModel().selectionChanged.connect(
self.commit
)
self.discRanksView.pressed.connect(self.onSelectItem)
self.discRanksView.horizontalHeader().sectionClicked.connect(
self.headerClick
)
self.discRanksView.verticalHeader().sectionClicked.connect(
self.onSelectItem
)
if self.headerState[0] is not None:
self.discRanksView.horizontalHeader().restoreState(
self.headerState[0])
self.contRanksView = QTableView()
self.ranksViewStack.addWidget(self.contRanksView)
self.contRanksView.setSelectionBehavior(QTableView.SelectRows)
self.contRanksView.setSelectionMode(QTableView.MultiSelection)
self.contRanksView.setSortingEnabled(True)
self.contRanksLabels = ["#"] + [m.shortname for m in self.contMeasures]
self.contRanksModel = QStandardItemModel(self)
self.contRanksModel.setHorizontalHeaderLabels(self.contRanksLabels)
self.contRanksProxyModel = MySortProxyModel(self)
self.contRanksProxyModel.setSourceModel(self.contRanksModel)
self.contRanksView.setModel(self.contRanksProxyModel)
self.contRanksView.setColumnWidth(0, 20)
self.contRanksView.selectionModel().selectionChanged.connect(
self.commit
)
self.contRanksView.pressed.connect(self.onSelectItem)
self.contRanksView.horizontalHeader().sectionClicked.connect(
self.headerClick
)
self.contRanksView.verticalHeader().sectionClicked.connect(
self.onSelectItem
)
if self.headerState[1] is not None:
self.contRanksView.horizontalHeader().restoreState(
self.headerState[1])
self.noClassRanksView = QTableView()
self.ranksViewStack.addWidget(self.noClassRanksView)
self.noClassRanksView.setSelectionBehavior(QTableView.SelectRows)
self.noClassRanksView.setSelectionMode(QTableView.MultiSelection)
self.noClassRanksView.setSortingEnabled(True)
self.noClassRanksLabels = ["#"]
self.noClassRanksModel = QStandardItemModel(self)
self.noClassRanksModel.setHorizontalHeaderLabels(self.noClassRanksLabels)
self.noClassRanksProxyModel = MySortProxyModel(self)
self.noClassRanksProxyModel.setSourceModel(self.noClassRanksModel)
self.noClassRanksView.setModel(self.noClassRanksProxyModel)
self.noClassRanksView.setColumnWidth(0, 20)
self.noClassRanksView.selectionModel().selectionChanged.connect(
self.commit
)
self.noClassRanksView.pressed.connect(self.onSelectItem)
self.noClassRanksView.horizontalHeader().sectionClicked.connect(
self.headerClick
)
self.noClassRanksView.verticalHeader().sectionClicked.connect(
self.onSelectItem
)
if self.headerState[2] is not None:
self.noClassRanksView.horizontalHeader().restoreState(
self.headerState[2])
# Switch the current view to Discrete
self.switchRanksMode(0)
self.resetInternals()
self.updateDelegates()
self.updateVisibleScoreColumns()
self.resize(690, 500)
self.measure_scores = table((len(self.measures), 0), None)
def switchRanksMode(self, index):
"""
Switch between discrete/continuous/no_class mode
"""
self.rankMode = index
self.ranksViewStack.setCurrentIndex(index)
if index == 0:
self.ranksView = self.discRanksView
self.ranksModel = self.discRanksModel
self.ranksProxyModel = self.discRanksProxyModel
self.measures = self.discMeasures
self.selected_checks = self.cls_default_selected
self.reg_scoring_box.setSizePolicy(QSizePolicy.Ignored,
QSizePolicy.Ignored)
self.cls_scoring_box.setSizePolicy(QSizePolicy.Expanding,
QSizePolicy.Expanding)
elif index == 1:
self.ranksView = self.contRanksView
self.ranksModel = self.contRanksModel
self.ranksProxyModel = self.contRanksProxyModel
self.measures = self.contMeasures
self.selected_checks = self.reg_default_selected
self.cls_scoring_box.setSizePolicy(QSizePolicy.Ignored,
QSizePolicy.Ignored)
self.reg_scoring_box.setSizePolicy(QSizePolicy.Expanding,
QSizePolicy.Expanding)
else:
self.ranksView = self.noClassRanksView
self.ranksModel = self.noClassRanksModel
self.ranksProxyModel = self.noClassRanksProxyModel
self.measures = []
self.selected_checks = set()
self.reg_scoring_box.setSizePolicy(QSizePolicy.Ignored,
QSizePolicy.Ignored)
self.cls_scoring_box.setSizePolicy(QSizePolicy.Ignored,
QSizePolicy.Ignored)
shape = (len(self.measures) + len(self.learners), 0)
self.measure_scores = table(shape, None)
self.update_scores = False
for check, score in zip(self.score_checks, SCORES):
check.setChecked(score.name in self.selected_checks)
self.update_scores = True
self.score_stack.setCurrentIndex(index)
self.updateVisibleScoreColumns()
@check_sql_input
def setData(self, data):
self.closeContext()
self.clear_messages()
self.resetInternals()
self.data = data
self.switchRanksMode(0)
if self.data is not None:
domain = self.data.domain
attrs = domain.attributes
self.usefulAttributes = [attr for attr in attrs
if attr.is_discrete or attr.is_continuous]
if domain.has_continuous_class:
self.switchRanksMode(1)
elif not domain.class_var:
self.Warning.no_target_var()
self.switchRanksMode(2)
elif not domain.has_discrete_class:
self.Error.invalid_type(type(domain.class_var).__name__)
if issparse(self.data.X): # keep only measures supporting sparse data
self.measures = [m for m in self.measures
if m.score.supports_sparse_data]
self.ranksModel.setRowCount(len(attrs))
for i, a in enumerate(attrs):
if a.is_discrete:
v = len(a.values)
else:
v = "C"
item = ScoreValueItem()
item.setData(v, Qt.DisplayRole)
self.ranksModel.setItem(i, 0, item)
item = QStandardItem(a.name)
item.setData(gui.attributeIconDict[a], Qt.DecorationRole)
self.ranksModel.setVerticalHeaderItem(i, item)
shape = (len(self.measures) + len(self.learners), len(attrs))
self.measure_scores = table(shape, None)
self.updateScores()
else:
self.send("Scores", None)
self.selected_rows = []
self.openContext(data)
self.selectMethodChanged()
self.commit()
def get_selection(self):
selection = self.ranksView.selectionModel().selection()
return list(set(ind.row() for ind in selection.indexes()))
def set_learner(self, learner, lid=None):
if learner is None and lid is not None:
del self.learners[lid]
elif learner is not None:
self.learners[lid] = score_meta(
learner.name,
learner.name,
learner
)
attrs_len = 0 if not self.data else len(self.data.domain.attributes)
shape = (len(self.learners), attrs_len)
self.measure_scores = self.measure_scores[:len(self.measures)]
self.measure_scores += table(shape, None)
self.contRanksModel.setHorizontalHeaderLabels(self.contRanksLabels)
self.discRanksModel.setHorizontalHeaderLabels(self.discRanksLabels)
self.noClassRanksModel.setHorizontalHeaderLabels(
self.noClassRanksLabels)
measures_mask = [False] * len(self.measures)
measures_mask += [True for _ in self.learners]
self.updateScores(measures_mask)
self.commit()
def updateScores(self, measuresMask=None):
"""
Update the current computed scores.
If `measuresMask` is given it must be an list of bool values
indicating what measures should be recomputed.
"""
if not self.data:
return
if self.data.has_missing():
self.information("Missing values have been imputed.")
measures = self.measures + [v for k, v in self.learners.items()]
if measuresMask is None:
# Update all selected measures
measuresMask = [self.selectedMeasures.get(m.name)
for m in self.measures]
measuresMask = measuresMask + [v.name for k, v in
self.learners.items()]
data = self.data
learner_col = len(self.measures)
if len(measuresMask) <= len(self.measures) or \
measuresMask[len(self.measures)]:
self.labels = []
self.Error.inadequate_learner.clear()
self.setStatusMessage("Running")
with self.progressBar():
n_measure_update = len([x for x in measuresMask if x is not False])
count = 0
for index, (meas, mask) in enumerate(zip(measures, measuresMask)):
if not mask:
continue
self.progressBarSet(90 * count / n_measure_update)
count += 1
if index < len(self.measures):
estimator = meas.score()
try:
self.measure_scores[index] = estimator(data)
except ValueError:
self.measure_scores[index] = []
for attr in data.domain.attributes:
try:
self.measure_scores[index].append(
estimator(data, attr))
except ValueError:
self.measure_scores[index].append(None)
else:
learner = meas.score
if isinstance(learner, Learner) and \
not learner.check_learner_adequacy(self.data.domain):
self.Error.inadequate_learner(
learner.learner_adequacy_err_msg)
scores = table((1, len(data.domain.attributes)))
else:
scores = meas.score.score_data(data)
for i, row in enumerate(scores):
self.labels.append(meas.shortname + str(i + 1))
if len(self.measure_scores) > learner_col:
self.measure_scores[learner_col] = row
else:
self.measure_scores.append(row)
learner_col += 1
self.progressBarSet(90)
self.contRanksModel.setHorizontalHeaderLabels(
self.contRanksLabels + self.labels
)
self.discRanksModel.setHorizontalHeaderLabels(
self.discRanksLabels + self.labels
)
self.noClassRanksModel.setHorizontalHeaderLabels(
self.noClassRanksLabels + self.labels
)
self.updateRankModel(measuresMask)
self.ranksProxyModel.invalidate()
self.selectMethodChanged()
self.send("Scores", self.create_scores_table(self.labels))
self.setStatusMessage("")
def updateRankModel(self, measuresMask):
"""
Update the rankModel.
"""
values = []
diff = len(self.measure_scores) - len(measuresMask)
if len(measuresMask):
measuresMask += [measuresMask[-1]] * diff
for i in range(self.ranksModel.columnCount() - 1,
len(self.measure_scores), -1):
self.ranksModel.removeColumn(i)
for i, (scores, m) in enumerate(zip(self.measure_scores, measuresMask)):
if not m and self.ranksModel.item(0, i + 1):
values.append([])
continue
values_one = []
for j, _score in enumerate(scores):
values_one.append(_score)
item = self.ranksModel.item(j, i + 1)
if not item:
item = ScoreValueItem()
self.ranksModel.setItem(j, i + 1, item)
item.setData(_score, Qt.DisplayRole)
values.append(values_one)
for i, (vals, m) in enumerate(zip(values, measuresMask)):
if not m:
continue
valid_vals = [v for v in vals if v is not None]
if valid_vals:
vmin, vmax = min(valid_vals), max(valid_vals)
for j, v in enumerate(vals):
if v is not None:
# Set the bar ratio role for i-th measure.
ratio = float((v - vmin) / ((vmax - vmin) or 1))
item = self.ranksModel.item(j, i + 1)
item.setData(ratio, gui.BarRatioRole)
self.ranksView.setColumnWidth(0, 20)
self.ranksView.resizeRowsToContents()
def resetInternals(self):
self.data = None
self.usefulAttributes = []
self.ranksModel.setRowCount(0)
def onSelectItem(self, index):
"""
Called when the user selects/unselects an item in the table view.
"""
self.selectMethod = OWRank.SelectManual # Manual
self.selectButtons.button(self.selectMethod).setChecked(True)
self.commit()
def setSelectMethod(self, method):
if self.selectMethod != method:
self.selectMethod = method
self.selectButtons.button(method).setChecked(True)
self.selectMethodChanged()
def selectMethodChanged(self):
self.autoSelection()
self.ranksView.setFocus()
def nSelectedChanged(self):
self.selectMethod = OWRank.SelectNBest
self.selectButtons.button(self.selectMethod).setChecked(True)
self.selectMethodChanged()
def autoSelection(self):
selModel = self.ranksView.selectionModel()
rowCount = self.ranksModel.rowCount()
columnCount = self.ranksModel.columnCount()
model = self.ranksProxyModel
if self.selectMethod == OWRank.SelectNone:
selection = QItemSelection()
elif self.selectMethod == OWRank.SelectAll:
selection = QItemSelection(
model.index(0, 0),
model.index(rowCount - 1, columnCount - 1)
)
elif self.selectMethod == OWRank.SelectNBest:
nSelected = min(self.nSelected, rowCount)
selection = QItemSelection(
model.index(0, 0),
model.index(nSelected - 1, columnCount - 1)
)
else:
selection = QItemSelection()
if len(self.selected_rows):
selection = QItemSelection()
for row in self.selected_rows:
selection.append(QItemSelectionRange(
model.index(row, 0), model.index(row, columnCount - 1)))
selModel.select(selection, QItemSelectionModel.ClearAndSelect)
def headerClick(self, index):
if index >= 1 and self.selectMethod == OWRank.SelectNBest:
# Reselect the top ranked attributes
self.autoSelection()
# Store the header states
disc = bytes(self.discRanksView.horizontalHeader().saveState())
cont = bytes(self.contRanksView.horizontalHeader().saveState())
no_class = bytes(self.noClassRanksView.horizontalHeader().saveState())
self.headerState = [disc, cont, no_class]
def measuresSelectionChanged(self, measure):
"""Measure selection has changed. Update column visibility.
"""
checked = self.selectedMeasures[measure.name]
self.selectedMeasures[measure.name] = not checked
if not checked:
self.selected_checks.add(measure.name)
elif measure.name in self.selected_checks:
self.selected_checks.remove(measure.name)
measures_mask = [False] * len(self.measures)
measures_mask += [False for _ in self.learners]
# Update scores for shown column if they are not yet computed.
if measure in self.measures and self.measure_scores:
index = self.measures.index(measure)
if all(s is None for s in self.measure_scores[index]):
measures_mask[index] = True
if self.update_scores:
self.updateScores(measures_mask)
self.updateVisibleScoreColumns()
def updateVisibleScoreColumns(self):
"""
Update the visible columns of the scores view.
"""
for i, measure in enumerate(self.measures):
shown = self.selectedMeasures.get(measure.name)
self.ranksView.setColumnHidden(i + 1, not shown)
self.ranksView.setColumnWidth(i + 1, 100)
index = self.ranksView.horizontalHeader().sortIndicatorSection()
if self.ranksView.isColumnHidden(index):
self.headerState[self.rankMode] = None
if self.headerState[self.rankMode] is None:
def get_sort_by_col(measures, selected_measures):
cols = [i + 1 for i, m in enumerate(measures) if
m.name in selected_measures]
return cols[0] if cols else len(measures) + 1
col = get_sort_by_col(self.measures, self.selected_checks)
self.ranksView.sortByColumn(col, Qt.DescendingOrder)
self.autoSelection()
def updateDelegates(self):
self.contRanksView.setItemDelegate(gui.ColoredBarItemDelegate(self))
self.discRanksView.setItemDelegate(gui.ColoredBarItemDelegate(self))
self.noClassRanksView.setItemDelegate(gui.ColoredBarItemDelegate(self))
def send_report(self):
if not self.data:
return
self.report_domain("Input", self.data.domain)
self.report_table("Ranks", self.ranksView, num_format="{:.3f}")
if self.out_domain_desc is not None:
self.report_items("Output", self.out_domain_desc)
def commit(self):
self.selected_rows = self.get_selection()
if self.data and len(self.data.domain.attributes) == len(
self.selected_rows):
self.selectMethod = OWRank.SelectAll
self.selectButtons.button(self.selectMethod).setChecked(True)
selected = self.selectedAttrs()
if not self.data or not selected:
self.send("Reduced Data", None)
self.out_domain_desc = None
else:
data = Table(Domain(selected, self.data.domain.class_var,
self.data.domain.metas), self.data)
self.send("Reduced Data", data)
self.out_domain_desc = report.describe_domain(data.domain)
def selectedAttrs(self):
if self.data:
inds = self.ranksView.selectionModel().selectedRows(0)
source = self.ranksProxyModel.mapToSource
inds = map(source, inds)
inds = [ind.row() for ind in inds]
return [self.data.domain.attributes[i] for i in inds]
else:
return []
def create_scores_table(self, labels):
indices = [i for i, m in enumerate(self.measures)
if self.selectedMeasures.get(m.name, False)]
measures = [s.name for s in self.measures if
self.selectedMeasures.get(s.name, False)]
measures += [label for label in labels]
if not measures:
return None
features = [ContinuousVariable(s) for s in measures]
metas = [StringVariable("Feature name")]
domain = Domain(features, metas=metas)
scores = np.array([row for i, row in enumerate(self.measure_scores)
if i in indices or i >= len(self.measures)]).T
feature_names = np.array([a.name for a in self.data.domain.attributes])
# Reshape to 2d array as Table does not like 1d arrays
feature_names = feature_names[:, None]
new_table = Table(domain, scores, metas=feature_names)
new_table.name = "Feature Scores"
return new_table
@classmethod
def migrate_settings(cls, settings, version):
if not version:
# Before fc5caa1e1d716607f1f5c4e0b0be265c23280fa0
# headerState had length 2
headerState = settings.get("headerState", None)
if headerState is not None and \
isinstance(headerState, tuple) and \
len(headerState) < 3:
headerState = (list(headerState) + [None] * 3)[:3]
settings["headerState"] = headerState
class OWRank(OWWidget):
name = "Rank"
description = "Rank and filter data features by their relevance."
icon = "icons/Rank.svg"
priority = 1102
keywords = []
buttons_area_orientation = Qt.Vertical
class Inputs:
data = Input("Data", Table)
scorer = Input("Scorer", score.Scorer, multiple=True)
class Outputs:
reduced_data = Output("Reduced Data", Table, default=True)
scores = Output("Scores", Table)
features = Output("Features", AttributeList, dynamic=False)
SelectNone, SelectAll, SelectManual, SelectNBest = range(4)
nSelected = ContextSetting(5)
auto_apply = Setting(True)
sorting = Setting((0, Qt.DescendingOrder))
selected_methods = Setting(set())
settings_version = 2
settingsHandler = DomainContextHandler()
selected_rows = ContextSetting([])
selectionMethod = ContextSetting(SelectNBest)
class Information(OWWidget.Information):
no_target_var = Msg("Data does not have a single target variable. "
"You can still connect in unsupervised scorers "
"such as PCA.")
missings_imputed = Msg('Missing values will be imputed as needed.')
class Error(OWWidget.Error):
invalid_type = Msg("Cannot handle target variable type {}")
inadequate_learner = Msg("Scorer {} inadequate: {}")
no_attributes = Msg("Data does not have a single attribute.")
def __init__(self):
super().__init__()
self.scorers = OrderedDict()
self.out_domain_desc = None
self.data = None
self.problem_type_mode = ProblemType.CLASSIFICATION
if not self.selected_methods:
self.selected_methods = {method.name for method in SCORES
if method.is_default}
# GUI
self.ranksModel = model = TableModel(parent=self) # type: TableModel
self.ranksView = view = TableView(self) # type: TableView
self.mainArea.layout().addWidget(view)
view.setModel(model)
view.setColumnWidth(0, 30)
view.selectionModel().selectionChanged.connect(self.on_select)
def _set_select_manual():
self.setSelectionMethod(OWRank.SelectManual)
view.pressed.connect(_set_select_manual)
view.verticalHeader().sectionClicked.connect(_set_select_manual)
view.horizontalHeader().sectionClicked.connect(self.headerClick)
self.measuresStack = stacked = QStackedWidget(self)
self.controlArea.layout().addWidget(stacked)
for scoring_methods in (CLS_SCORES,
REG_SCORES,
[]):
box = gui.vBox(None, "Scoring Methods" if scoring_methods else None)
stacked.addWidget(box)
for method in scoring_methods:
box.layout().addWidget(QCheckBox(
method.name, self,
objectName=method.shortname, # To be easily found in tests
checked=method.name in self.selected_methods,
stateChanged=partial(self.methodSelectionChanged, method_name=method.name)))
gui.rubber(box)
gui.rubber(self.controlArea)
self.switchProblemType(ProblemType.CLASSIFICATION)
selMethBox = gui.vBox(self.controlArea, "Select Attributes", addSpace=True)
grid = QGridLayout()
grid.setContentsMargins(6, 0, 6, 0)
self.selectButtons = QButtonGroup()
self.selectButtons.buttonClicked[int].connect(self.setSelectionMethod)
def button(text, buttonid, toolTip=None):
b = QRadioButton(text)
self.selectButtons.addButton(b, buttonid)
if toolTip is not None:
b.setToolTip(toolTip)
return b
b1 = button(self.tr("None"), OWRank.SelectNone)
b2 = button(self.tr("All"), OWRank.SelectAll)
b3 = button(self.tr("Manual"), OWRank.SelectManual)
b4 = button(self.tr("Best ranked:"), OWRank.SelectNBest)
s = gui.spin(selMethBox, self, "nSelected", 1, 100,
callback=lambda: self.setSelectionMethod(OWRank.SelectNBest))
grid.addWidget(b1, 0, 0)
grid.addWidget(b2, 1, 0)
grid.addWidget(b3, 2, 0)
grid.addWidget(b4, 3, 0)
grid.addWidget(s, 3, 1)
self.selectButtons.button(self.selectionMethod).setChecked(True)
selMethBox.layout().addLayout(grid)
gui.auto_commit(selMethBox, self, "auto_apply", "Send", box=False)
self.resize(690, 500)
def switchProblemType(self, index):
"""
Switch between discrete/continuous/no_class mode
"""
self.measuresStack.setCurrentIndex(index)
self.problem_type_mode = index
@Inputs.data
@check_sql_input
def set_data(self, data):
self.closeContext()
self.selected_rows = []
self.ranksModel.clear()
self.ranksModel.resetSorting(True)
self.get_method_scores.cache_clear()
self.get_scorer_scores.cache_clear()
self.Error.clear()
self.Information.clear()
self.Information.missings_imputed(
shown=data is not None and data.has_missing())
if data is not None and not len(data.domain.attributes):
data = None
self.Error.no_attributes()
self.data = data
self.switchProblemType(ProblemType.CLASSIFICATION)
if self.data is not None:
domain = self.data.domain
if domain.has_discrete_class:
problem_type = ProblemType.CLASSIFICATION
elif domain.has_continuous_class:
problem_type = ProblemType.REGRESSION
elif not domain.class_var:
self.Information.no_target_var()
problem_type = ProblemType.UNSUPERVISED
else:
# This can happen?
self.Error.invalid_type(type(domain.class_var).__name__)
problem_type = None
if problem_type is not None:
self.switchProblemType(problem_type)
self.ranksModel.setVerticalHeaderLabels(domain.attributes)
self.ranksView.setVHeaderFixedWidthFromLabel(
max((a.name for a in domain.attributes), key=len))
self.selectionMethod = OWRank.SelectNBest
self.openContext(data)
self.selectButtons.button(self.selectionMethod).setChecked(True)
def handleNewSignals(self):
self.setStatusMessage('Running')
self.updateScores()
self.setStatusMessage('')
self.on_select()
@Inputs.scorer
def set_learner(self, scorer, id):
if scorer is None:
self.scorers.pop(id, None)
else:
# Avoid caching a (possibly stale) previous instance of the same
# Scorer passed via the same signal
if id in self.scorers:
self.get_scorer_scores.cache_clear()
self.scorers[id] = ScoreMeta(scorer.name, scorer.name, scorer,
ProblemType.from_variable(scorer.class_type),
False)
@memoize_method()
def get_method_scores(self, method):
estimator = method.scorer()
data = self.data
try:
scores = np.asarray(estimator(data))
except ValueError:
log.warning("Scorer %s wasn't able to compute all scores at once",
method.name)
try:
scores = np.array([estimator(data, attr)
for attr in data.domain.attributes])
except ValueError:
log.error(
"Scorer %s wasn't able to compute scores at all",
method.name)
scores = np.full(len(data.domain.attributes), np.nan)
return scores
@memoize_method()
def get_scorer_scores(self, scorer):
try:
scores = scorer.scorer.score_data(self.data).T
except ValueError:
log.error(
"Scorer %s wasn't able to compute scores at all",
scorer.name)
scores = np.full((len(self.data.domain.attributes), 1), np.nan)
labels = ((scorer.shortname,)
if scores.shape[1] == 1 else
tuple(scorer.shortname + '_' + str(i)
for i in range(1, 1 + scores.shape[1])))
return scores, labels
def updateScores(self):
if self.data is None:
self.ranksModel.clear()
self.Outputs.scores.send(None)
return
methods = [method
for method in SCORES
if (method.name in self.selected_methods and
method.problem_type == self.problem_type_mode and
(not issparse(self.data.X) or
method.scorer.supports_sparse_data))]
scorers = []
self.Error.inadequate_learner.clear()
for scorer in self.scorers.values():
if scorer.problem_type in (self.problem_type_mode, ProblemType.UNSUPERVISED):
scorers.append(scorer)
else:
self.Error.inadequate_learner(scorer.name, scorer.learner_adequacy_err_msg)
method_scores = tuple(self.get_method_scores(method)
for method in methods)
scorer_scores, scorer_labels = (), ()
if scorers:
scorer_scores, scorer_labels = zip(*(self.get_scorer_scores(scorer)
for scorer in scorers))
scorer_labels = tuple(chain.from_iterable(scorer_labels))
labels = tuple(method.shortname for method in methods) + scorer_labels
model_array = np.column_stack(
([len(a.values) if a.is_discrete else np.nan
for a in self.data.domain.attributes],) +
(method_scores if method_scores else ()) +
(scorer_scores if scorer_scores else ())
)
for column, values in enumerate(model_array.T):
self.ranksModel.setExtremesFrom(column, values)
self.ranksModel.wrap(model_array.tolist())
self.ranksModel.setHorizontalHeaderLabels(('#',) + labels)
self.ranksView.setColumnWidth(0, 40)
# Re-apply sort
try:
sort_column, sort_order = self.sorting
if sort_column < len(labels):
# adds 1 for '#' (discrete count) column
self.ranksModel.sort(sort_column + 1, sort_order)
self.ranksView.horizontalHeader().setSortIndicator(sort_column + 1, sort_order)
except ValueError:
pass
self.autoSelection()
self.Outputs.scores.send(self.create_scores_table(labels))
def on_select(self):
# Save indices of attributes in the original, unsorted domain
self.selected_rows = self.ranksModel.mapToSourceRows([
i.row() for i in self.ranksView.selectionModel().selectedRows(0)])
self.commit()
def setSelectionMethod(self, method):
self.selectionMethod = method
self.selectButtons.button(method).setChecked(True)
self.autoSelection()
def autoSelection(self):
selModel = self.ranksView.selectionModel()
model = self.ranksModel
rowCount = model.rowCount()
columnCount = model.columnCount()
if self.selectionMethod == OWRank.SelectNone:
selection = QItemSelection()
elif self.selectionMethod == OWRank.SelectAll:
selection = QItemSelection(
model.index(0, 0),
model.index(rowCount - 1, columnCount - 1)
)
elif self.selectionMethod == OWRank.SelectNBest:
nSelected = min(self.nSelected, rowCount)
selection = QItemSelection(
model.index(0, 0),
model.index(nSelected - 1, columnCount - 1)
)
else:
selection = QItemSelection()
if len(self.selected_rows):
for row in model.mapFromSourceRows(self.selected_rows):
selection.append(QItemSelectionRange(
model.index(row, 0), model.index(row, columnCount - 1)))
selModel.select(selection, QItemSelectionModel.ClearAndSelect)
def headerClick(self, index):
if index >= 1 and self.selectionMethod == OWRank.SelectNBest:
# Reselect the top ranked attributes
self.autoSelection()
# Store the header states
sort_order = self.ranksModel.sortOrder()
sort_column = self.ranksModel.sortColumn() - 1 # -1 for '#' (discrete count) column
self.sorting = (sort_column, sort_order)
def methodSelectionChanged(self, state, method_name):
if state == Qt.Checked:
self.selected_methods.add(method_name)
elif method_name in self.selected_methods:
self.selected_methods.remove(method_name)
self.updateScores()
def send_report(self):
if not self.data:
return
self.report_domain("Input", self.data.domain)
self.report_table("Ranks", self.ranksView, num_format="{:.3f}")
if self.out_domain_desc is not None:
self.report_items("Output", self.out_domain_desc)
def commit(self):
selected_attrs = []
if self.data is not None:
selected_attrs = [self.data.domain.attributes[i]
for i in self.selected_rows]
if not selected_attrs:
self.Outputs.reduced_data.send(None)
self.Outputs.features.send(None)
self.out_domain_desc = None
else:
reduced_domain = Domain(
selected_attrs, self.data.domain.class_var, self.data.domain.metas)
data = self.data.transform(reduced_domain)
self.Outputs.reduced_data.send(data)
self.Outputs.features.send(AttributeList(selected_attrs))
self.out_domain_desc = report.describe_domain(data.domain)
def create_scores_table(self, labels):
model_list = self.ranksModel.tolist()
if not model_list or len(model_list[0]) == 1: # Empty or just n_values column
return None
domain = Domain([ContinuousVariable(label) for label in labels],
metas=[StringVariable("Feature")])
# Prevent np.inf scores
finfo = np.finfo(np.float64)
scores = np.clip(np.array(model_list)[:, 1:], finfo.min, finfo.max)
feature_names = np.array([a.name for a in self.data.domain.attributes])
# Reshape to 2d array as Table does not like 1d arrays
feature_names = feature_names[:, None]
new_table = Table(domain, scores, metas=feature_names)
new_table.name = "Feature Scores"
return new_table
@classmethod
def migrate_settings(cls, settings, version):
# If older settings, restore sort header to default
# Saved selected_rows will likely be incorrect
if version is None or version < 2:
column, order = 0, Qt.DescendingOrder
headerState = settings.pop("headerState", None)
# Lacking knowledge of last problemType, use discrete ranks view's ordering
if isinstance(headerState, (tuple, list)):
headerState = headerState[0]
if isinstance(headerState, bytes):
hview = QHeaderView(Qt.Horizontal)
hview.restoreState(headerState)
column, order = hview.sortIndicatorSection() - 1, hview.sortIndicatorOrder()
settings["sorting"] = (column, order)
@classmethod
def migrate_context(cls, context, version):
if version is None or version < 2:
# Old selection was saved as sorted indices. New selection is original indices.
# Since we can't devise the latter without first computing the ranks,
# just reset the selection to avoid confusion.
context.values['selected_rows'] = []
class OWKeywords(OWWidget, ConcurrentWidgetMixin):
name = "Extract Keywords"
description = "Infers characteristic words from the input corpus."
icon = "icons/Keywords.svg"
priority = 1100
keywords = ["characteristic", "term"]
DEFAULT_SORTING = (1, Qt.DescendingOrder)
settingsHandler = DomainContextHandler()
selected_scoring_methods: Set[str] = Setting({ScoringMethods.TF_IDF})
yake_lang_index: int = Setting(YAKE_LANGUAGES.index("English"))
rake_lang_index: int = Setting(RAKE_LANGUAGES.index("English"))
agg_method: int = Setting(AggregationMethods.MEAN)
sel_method: int = ContextSetting(SelectionMethods.N_BEST)
n_selected: int = ContextSetting(3)
sort_column_order: Tuple[int, int] = Setting(DEFAULT_SORTING)
selected_words = ContextSetting([], schema_only=True)
auto_apply: bool = Setting(True)
class Inputs:
corpus = Input("Corpus", Corpus)
words = Input("Words", Table)
class Outputs:
words = Output("Words", Corpus)
class Warning(OWWidget.Warning):
no_words_column = Msg("Input is missing 'Words' column.")
def __init__(self):
OWWidget.__init__(self)
ConcurrentWidgetMixin.__init__(self)
self.corpus: Optional[Corpus] = None
self.words: Optional[List] = None
self.__cached_keywords = {}
self.model = KeywordsTableModel(parent=self)
self._setup_gui()
def _setup_gui(self):
grid = QGridLayout()
box = gui.widgetBox(self.controlArea, "Scoring Methods", grid)
yake_cb = gui.comboBox(self.controlArea,
self,
"yake_lang_index",
items=YAKE_LANGUAGES,
callback=self.__on_yake_lang_changed)
rake_cb = gui.comboBox(self.controlArea,
self,
"rake_lang_index",
items=RAKE_LANGUAGES,
callback=self.__on_rake_lang_changed)
for i, (method_name, _) in enumerate(ScoringMethods.ITEMS):
check_box = QCheckBox(method_name, self)
check_box.setChecked(method_name in self.selected_scoring_methods)
check_box.stateChanged.connect(
lambda state, name=method_name: self.
__on_scoring_method_state_changed(state, name))
box.layout().addWidget(check_box, i, 0)
if method_name == ScoringMethods.YAKE:
box.layout().addWidget(yake_cb, i, 1)
if method_name == ScoringMethods.RAKE:
box.layout().addWidget(rake_cb, i, 1)
box = gui.vBox(self.controlArea, "Aggregation")
gui.comboBox(box,
self,
"agg_method",
items=AggregationMethods.ITEMS,
callback=self.update_scores)
box = gui.vBox(self.controlArea, "Select Words")
grid = QGridLayout()
grid.setContentsMargins(0, 0, 0, 0)
box.layout().addLayout(grid)
self.__sel_method_buttons = QButtonGroup()
for method, label in enumerate(SelectionMethods.ITEMS):
button = QRadioButton(label)
button.setChecked(method == self.sel_method)
grid.addWidget(button, method, 0)
self.__sel_method_buttons.addButton(button, method)
self.__sel_method_buttons.buttonClicked[int].connect(
self._set_selection_method)
spin = gui.spin(box,
self,
"n_selected",
1,
999,
addToLayout=False,
callback=lambda: self._set_selection_method(
SelectionMethods.N_BEST))
grid.addWidget(spin, 3, 1)
gui.rubber(self.controlArea)
gui.auto_send(self.buttonsArea, self, "auto_apply")
self.__filter_line_edit = QLineEdit(
textChanged=self.__on_filter_changed, placeholderText="Filter...")
self.mainArea.layout().addWidget(self.__filter_line_edit)
def select_manual():
self._set_selection_method(SelectionMethods.MANUAL)
self.view = KeywordsTableView()
self.view.pressedAny.connect(select_manual)
self.view.horizontalHeader().setSortIndicator(*self.DEFAULT_SORTING)
self.view.horizontalHeader().sectionClicked.connect(
self.__on_horizontal_header_clicked)
self.mainArea.layout().addWidget(self.view)
proxy = SortFilterProxyModel()
proxy.setFilterKeyColumn(0)
proxy.setFilterCaseSensitivity(False)
self.view.setModel(proxy)
self.view.model().setSourceModel(self.model)
self.view.selectionModel().selectionChanged.connect(
self.__on_selection_changed)
def __on_scoring_method_state_changed(self, state: int, method_name: str):
if state == Qt.Checked:
self.selected_scoring_methods.add(method_name)
elif method_name in self.selected_scoring_methods:
self.selected_scoring_methods.remove(method_name)
self.update_scores()
def __on_yake_lang_changed(self):
if ScoringMethods.YAKE in self.selected_scoring_methods:
if ScoringMethods.YAKE in self.__cached_keywords:
del self.__cached_keywords[ScoringMethods.YAKE]
self.update_scores()
def __on_rake_lang_changed(self):
if ScoringMethods.RAKE in self.selected_scoring_methods:
if ScoringMethods.RAKE in self.__cached_keywords:
del self.__cached_keywords[ScoringMethods.RAKE]
self.update_scores()
def __on_filter_changed(self):
model = self.view.model()
model.setFilterFixedString(self.__filter_line_edit.text().strip())
self._select_rows()
def __on_horizontal_header_clicked(self, index: int):
header = self.view.horizontalHeader()
self.sort_column_order = (index, header.sortIndicatorOrder())
self._select_rows()
# explicitly call commit, because __on_selection_changed will not be
# invoked, since selection is actually the same, only order is not
if self.sel_method == SelectionMethods.MANUAL and self.selected_words \
or self.sel_method == SelectionMethods.ALL:
self.commit()
def __on_selection_changed(self):
selected_rows = self.view.selectionModel().selectedRows(0)
model = self.view.model()
self.selected_words = [
model.data(model.index(i.row(), 0)) for i in selected_rows
]
self.commit()
@Inputs.corpus
def set_corpus(self, corpus: Optional[Corpus]):
self.closeContext()
self._clear()
self.corpus = corpus
self.openContext(self.corpus)
self.__sel_method_buttons.button(self.sel_method).setChecked(True)
def _clear(self):
self.clear_messages()
self.cancel()
self.selected_words = []
self.model.clear()
self.__cached_keywords = {}
@Inputs.words
def set_words(self, words: Optional[Table]):
self.words = None
self.Warning.no_words_column.clear()
if words:
if WORDS_COLUMN_NAME in words.domain and words.domain[
WORDS_COLUMN_NAME].attributes.get("type") == "words":
self.words = list(words.get_column_view(WORDS_COLUMN_NAME)[0])
else:
self.Warning.no_words_column()
def handleNewSignals(self):
self.update_scores()
def update_scores(self):
kwargs = {
ScoringMethods.YAKE: {
"language": YAKE_LANGUAGES[self.yake_lang_index],
"max_len": self.corpus.ngram_range[1] if self.corpus else 1
},
ScoringMethods.RAKE: {
"language": RAKE_LANGUAGES[self.rake_lang_index],
"max_len": self.corpus.ngram_range[1] if self.corpus else 1
},
}
self.start(run, self.corpus, self.words, self.__cached_keywords,
self.selected_scoring_methods, kwargs, self.agg_method)
def _set_selection_method(self, method: int):
self.sel_method = method
self.__sel_method_buttons.button(method).setChecked(True)
self._select_rows()
def _select_rows(self):
model = self.view.model()
n_rows, n_columns = model.rowCount(), model.columnCount()
if self.sel_method == SelectionMethods.NONE:
selection = QItemSelection()
elif self.sel_method == SelectionMethods.ALL:
selection = QItemSelection(model.index(0, 0),
model.index(n_rows - 1, n_columns - 1))
elif self.sel_method == SelectionMethods.MANUAL:
selection = QItemSelection()
for i in range(n_rows):
word = model.data(model.index(i, 0))
if word in self.selected_words:
_selection = QItemSelection(model.index(i, 0),
model.index(i, n_columns - 1))
selection.merge(_selection, QItemSelectionModel.Select)
elif self.sel_method == SelectionMethods.N_BEST:
n_sel = min(self.n_selected, n_rows)
selection = QItemSelection(model.index(0, 0),
model.index(n_sel - 1, n_columns - 1))
else:
raise NotImplementedError
self.view.selectionModel().select(selection,
QItemSelectionModel.ClearAndSelect)
def on_exception(self, ex: Exception):
raise ex
def on_partial_result(self, _: Any):
pass
# pylint: disable=arguments-differ
def on_done(self, results: Results):
self.__cached_keywords = results.all_keywords
self.model.wrap(results.scores)
self.model.setHorizontalHeaderLabels(["Word"] + results.labels)
self._apply_sorting()
if self.model.rowCount() > 0:
self._select_rows()
else:
self.__on_selection_changed()
def _apply_sorting(self):
if self.model.columnCount() <= self.sort_column_order[0]:
self.sort_column_order = self.DEFAULT_SORTING
header = self.view.horizontalHeader()
current_sorting = (header.sortIndicatorSection(),
header.sortIndicatorOrder())
if current_sorting != self.sort_column_order:
header.setSortIndicator(*self.sort_column_order)
# needed to sort nans; 1. column has strings
# if self.sort_column_order[0] > 0:
# self.model.sort(*self.sort_column_order)
def onDeleteWidget(self):
self.shutdown()
super().onDeleteWidget()
def commit(self):
words = None
if self.selected_words:
words_var = StringVariable(WORDS_COLUMN_NAME)
words_var.attributes = {"type": "words"}
model = self.model
attrs = [
ContinuousVariable(model.headerData(i + 1, Qt.Horizontal))
for i in range(len(self.selected_scoring_methods))
]
domain = Domain(attrs, metas=[words_var])
sort_column, reverse = self.sort_column_order
data = sorted(model, key=lambda a: a[sort_column], reverse=reverse)
data = [s[1:] + s[:1] for s in data if s[0] in self.selected_words]
words = Table.from_list(domain, data)
words.name = "Words"
self.Outputs.words.send(words)
def send_report(self):
if not self.corpus:
return
self.report_data("Corpus", self.corpus)
if self.words is not None:
self.report_paragraph("Words", ", ".join(self.words))
self.report_table("Keywords", self.view, num_format="{:.3f}")
class OWRank(OWWidget):
name = "Rank"
description = "Rank and filter data features by their relevance."
icon = "icons/Rank.svg"
priority = 1102
keywords = []
buttons_area_orientation = Qt.Vertical
class Inputs:
data = Input("Data", Table)
scorer = Input("Scorer", score.Scorer, multiple=True)
class Outputs:
reduced_data = Output("Reduced Data", Table, default=True)
scores = Output("Scores", Table)
features = Output("Features", AttributeList, dynamic=False)
SelectNone, SelectAll, SelectManual, SelectNBest = range(4)
nSelected = ContextSetting(5)
auto_apply = Setting(True)
sorting = Setting((0, Qt.DescendingOrder))
selected_methods = Setting(set())
settings_version = 2
settingsHandler = DomainContextHandler()
selected_rows = ContextSetting([])
selectionMethod = ContextSetting(SelectNBest)
class Information(OWWidget.Information):
no_target_var = Msg("Data does not have a single target variable. "
"You can still connect in unsupervised scorers "
"such as PCA.")
missings_imputed = Msg('Missing values will be imputed as needed.')
class Error(OWWidget.Error):
invalid_type = Msg("Cannot handle target variable type {}")
inadequate_learner = Msg("Scorer {} inadequate: {}")
no_attributes = Msg("Data does not have a single attribute.")
def __init__(self):
super().__init__()
self.scorers = OrderedDict()
self.out_domain_desc = None
self.data = None
self.problem_type_mode = ProblemType.CLASSIFICATION
if not self.selected_methods:
self.selected_methods = {method.name for method in SCORES
if method.is_default}
# GUI
self.ranksModel = model = TableModel(parent=self) # type: TableModel
self.ranksView = view = TableView(self) # type: TableView
self.mainArea.layout().addWidget(view)
view.setModel(model)
view.setColumnWidth(0, 30)
view.selectionModel().selectionChanged.connect(self.on_select)
def _set_select_manual():
self.setSelectionMethod(OWRank.SelectManual)
view.pressed.connect(_set_select_manual)
view.verticalHeader().sectionClicked.connect(_set_select_manual)
view.horizontalHeader().sectionClicked.connect(self.headerClick)
self.measuresStack = stacked = QStackedWidget(self)
self.controlArea.layout().addWidget(stacked)
for scoring_methods in (CLS_SCORES,
REG_SCORES,
[]):
box = gui.vBox(None, "Scoring Methods" if scoring_methods else None)
stacked.addWidget(box)
for method in scoring_methods:
box.layout().addWidget(QCheckBox(
method.name, self,
objectName=method.shortname, # To be easily found in tests
checked=method.name in self.selected_methods,
stateChanged=partial(self.methodSelectionChanged, method_name=method.name)))
gui.rubber(box)
gui.rubber(self.controlArea)
self.switchProblemType(ProblemType.CLASSIFICATION)
selMethBox = gui.vBox(self.controlArea, "Select Attributes", addSpace=True)
grid = QGridLayout()
grid.setContentsMargins(6, 0, 6, 0)
self.selectButtons = QButtonGroup()
self.selectButtons.buttonClicked[int].connect(self.setSelectionMethod)
def button(text, buttonid, toolTip=None):
b = QRadioButton(text)
self.selectButtons.addButton(b, buttonid)
if toolTip is not None:
b.setToolTip(toolTip)
return b
b1 = button(self.tr("None"), OWRank.SelectNone)
b2 = button(self.tr("All"), OWRank.SelectAll)
b3 = button(self.tr("Manual"), OWRank.SelectManual)
b4 = button(self.tr("Best ranked:"), OWRank.SelectNBest)
s = gui.spin(selMethBox, self, "nSelected", 1, 100,
callback=lambda: self.setSelectionMethod(OWRank.SelectNBest))
grid.addWidget(b1, 0, 0)
grid.addWidget(b2, 1, 0)
grid.addWidget(b3, 2, 0)
grid.addWidget(b4, 3, 0)
grid.addWidget(s, 3, 1)
self.selectButtons.button(self.selectionMethod).setChecked(True)
selMethBox.layout().addLayout(grid)
gui.auto_commit(selMethBox, self, "auto_apply", "Send", box=False)
self.resize(690, 500)
def switchProblemType(self, index):
"""
Switch between discrete/continuous/no_class mode
"""
self.measuresStack.setCurrentIndex(index)
self.problem_type_mode = index
@Inputs.data
@check_sql_input
def set_data(self, data):
self.closeContext()
self.selected_rows = []
self.ranksModel.clear()
self.ranksModel.resetSorting(True)
self.get_method_scores.cache_clear()
self.get_scorer_scores.cache_clear()
self.Error.clear()
self.Information.clear()
self.Information.missings_imputed(
shown=data is not None and data.has_missing())
if data is not None and not len(data.domain.attributes):
data = None
self.Error.no_attributes()
self.data = data
self.switchProblemType(ProblemType.CLASSIFICATION)
if self.data is not None:
domain = self.data.domain
if domain.has_discrete_class:
problem_type = ProblemType.CLASSIFICATION
elif domain.has_continuous_class:
problem_type = ProblemType.REGRESSION
elif not domain.class_var:
self.Information.no_target_var()
problem_type = ProblemType.UNSUPERVISED
else:
# This can happen?
self.Error.invalid_type(type(domain.class_var).__name__)
problem_type = None
if problem_type is not None:
self.switchProblemType(problem_type)
self.ranksModel.setVerticalHeaderLabels(domain.attributes)
self.ranksView.setVHeaderFixedWidthFromLabel(
max((a.name for a in domain.attributes), key=len))
self.selectionMethod = OWRank.SelectNBest
self.openContext(data)
self.selectButtons.button(self.selectionMethod).setChecked(True)
def handleNewSignals(self):
self.setStatusMessage('Running')
self.updateScores()
self.setStatusMessage('')
self.on_select()
@Inputs.scorer
def set_learner(self, scorer, id):
if scorer is None:
self.scorers.pop(id, None)
else:
# Avoid caching a (possibly stale) previous instance of the same
# Scorer passed via the same signal
if id in self.scorers:
self.get_scorer_scores.cache_clear()
self.scorers[id] = ScoreMeta(scorer.name, scorer.name, scorer,
ProblemType.from_variable(scorer.class_type),
False)
@memoize_method()
def get_method_scores(self, method):
estimator = method.scorer()
data = self.data
try:
scores = np.asarray(estimator(data))
except ValueError:
log.warning("Scorer %s wasn't able to compute all scores at once",
method.name)
try:
scores = np.array([estimator(data, attr)
for attr in data.domain.attributes])
except ValueError:
log.error(
"Scorer %s wasn't able to compute scores at all",
method.name)
scores = np.full(len(data.domain.attributes), np.nan)
return scores
@memoize_method()
def get_scorer_scores(self, scorer):
try:
scores = scorer.scorer.score_data(self.data).T
except ValueError:
log.error(
"Scorer %s wasn't able to compute scores at all",
scorer.name)
scores = np.full((len(self.data.domain.attributes), 1), np.nan)
labels = ((scorer.shortname,)
if scores.shape[1] == 1 else
tuple(scorer.shortname + '_' + str(i)
for i in range(1, 1 + scores.shape[1])))
return scores, labels
def updateScores(self):
if self.data is None:
self.ranksModel.clear()
self.Outputs.scores.send(None)
return
methods = [method
for method in SCORES
if (method.name in self.selected_methods and
method.problem_type == self.problem_type_mode and
(not issparse(self.data.X) or
method.scorer.supports_sparse_data))]
scorers = []
self.Error.inadequate_learner.clear()
for scorer in self.scorers.values():
if scorer.problem_type in (self.problem_type_mode, ProblemType.UNSUPERVISED):
scorers.append(scorer)
else:
self.Error.inadequate_learner(scorer.name, scorer.learner_adequacy_err_msg)
method_scores = tuple(self.get_method_scores(method)
for method in methods)
scorer_scores, scorer_labels = (), ()
if scorers:
scorer_scores, scorer_labels = zip(*(self.get_scorer_scores(scorer)
for scorer in scorers))
scorer_labels = tuple(chain.from_iterable(scorer_labels))
labels = tuple(method.shortname for method in methods) + scorer_labels
model_array = np.column_stack(
([len(a.values) if a.is_discrete else np.nan
for a in self.data.domain.attributes],) +
(method_scores if method_scores else ()) +
(scorer_scores if scorer_scores else ())
)
for column, values in enumerate(model_array.T):
self.ranksModel.setExtremesFrom(column, values)
self.ranksModel.wrap(model_array.tolist())
self.ranksModel.setHorizontalHeaderLabels(('#',) + labels)
self.ranksView.setColumnWidth(0, 40)
# Re-apply sort
try:
sort_column, sort_order = self.sorting
if sort_column < len(labels):
# adds 1 for '#' (discrete count) column
self.ranksModel.sort(sort_column + 1, sort_order)
self.ranksView.horizontalHeader().setSortIndicator(sort_column + 1, sort_order)
except ValueError:
pass
self.autoSelection()
self.Outputs.scores.send(self.create_scores_table(labels))
def on_select(self):
# Save indices of attributes in the original, unsorted domain
self.selected_rows = self.ranksModel.mapToSourceRows([
i.row() for i in self.ranksView.selectionModel().selectedRows(0)])
self.commit()
def setSelectionMethod(self, method):
self.selectionMethod = method
self.selectButtons.button(method).setChecked(True)
self.autoSelection()
def autoSelection(self):
selModel = self.ranksView.selectionModel()
model = self.ranksModel
rowCount = model.rowCount()
columnCount = model.columnCount()
if self.selectionMethod == OWRank.SelectNone:
selection = QItemSelection()
elif self.selectionMethod == OWRank.SelectAll:
selection = QItemSelection(
model.index(0, 0),
model.index(rowCount - 1, columnCount - 1)
)
elif self.selectionMethod == OWRank.SelectNBest:
nSelected = min(self.nSelected, rowCount)
selection = QItemSelection(
model.index(0, 0),
model.index(nSelected - 1, columnCount - 1)
)
else:
selection = QItemSelection()
if len(self.selected_rows):
for row in model.mapFromSourceRows(self.selected_rows):
selection.append(QItemSelectionRange(
model.index(row, 0), model.index(row, columnCount - 1)))
selModel.select(selection, QItemSelectionModel.ClearAndSelect)
def headerClick(self, index):
if index >= 1 and self.selectionMethod == OWRank.SelectNBest:
# Reselect the top ranked attributes
self.autoSelection()
# Store the header states
sort_order = self.ranksModel.sortOrder()
sort_column = self.ranksModel.sortColumn() - 1 # -1 for '#' (discrete count) column
self.sorting = (sort_column, sort_order)
def methodSelectionChanged(self, state, method_name):
if state == Qt.Checked:
self.selected_methods.add(method_name)
elif method_name in self.selected_methods:
self.selected_methods.remove(method_name)
self.updateScores()
def send_report(self):
if not self.data:
return
self.report_domain("Input", self.data.domain)
self.report_table("Ranks", self.ranksView, num_format="{:.3f}")
if self.out_domain_desc is not None:
self.report_items("Output", self.out_domain_desc)
def commit(self):
selected_attrs = []
if self.data is not None:
selected_attrs = [self.data.domain.attributes[i]
for i in self.selected_rows]
if not selected_attrs:
self.Outputs.reduced_data.send(None)
self.Outputs.features.send(None)
self.out_domain_desc = None
else:
reduced_domain = Domain(
selected_attrs, self.data.domain.class_var, self.data.domain.metas)
data = self.data.transform(reduced_domain)
self.Outputs.reduced_data.send(data)
self.Outputs.features.send(AttributeList(selected_attrs))
self.out_domain_desc = report.describe_domain(data.domain)
def create_scores_table(self, labels):
model_list = self.ranksModel.tolist()
if not model_list or len(model_list[0]) == 1: # Empty or just n_values column
return None
domain = Domain([ContinuousVariable(label) for label in labels],
metas=[StringVariable("Feature")])
# Prevent np.inf scores
finfo = np.finfo(np.float64)
scores = np.clip(np.array(model_list)[:, 1:], finfo.min, finfo.max)
feature_names = np.array([a.name for a in self.data.domain.attributes])
# Reshape to 2d array as Table does not like 1d arrays
feature_names = feature_names[:, None]
new_table = Table(domain, scores, metas=feature_names)
new_table.name = "Feature Scores"
return new_table
@classmethod
def migrate_settings(cls, settings, version):
# If older settings, restore sort header to default
# Saved selected_rows will likely be incorrect
if version is None or version < 2:
column, order = 0, Qt.DescendingOrder
headerState = settings.pop("headerState", None)
# Lacking knowledge of last problemType, use discrete ranks view's ordering
if isinstance(headerState, (tuple, list)):
headerState = headerState[0]
if isinstance(headerState, bytes):
hview = QHeaderView(Qt.Horizontal)
hview.restoreState(headerState)
column, order = hview.sortIndicatorSection() - 1, hview.sortIndicatorOrder()
settings["sorting"] = (column, order)
@classmethod
def migrate_context(cls, context, version):
if version is None or version < 2:
# Old selection was saved as sorted indices. New selection is original indices.
# Since we can't devise the latter without first computing the ranks,
# just reset the selection to avoid confusion.
context.values['selected_rows'] = []
class OWScoreDocuments(OWWidget, ConcurrentWidgetMixin):
name = "Score Documents"
description = ""
icon = "icons/ScoreDocuments.svg"
priority = 500
buttons_area_orientation = Qt.Vertical
# default order - table sorted in input order
DEFAULT_SORTING = (-1, Qt.AscendingOrder)
settingsHandler = PerfectDomainContextHandler()
auto_commit: bool = Setting(True)
aggregation: int = Setting(0)
word_frequency: bool = Setting(True)
word_appearance: bool = Setting(False)
embedding_similarity: bool = Setting(False)
embedding_language: int = Setting(0)
sort_column_order: Tuple[int, int] = Setting(DEFAULT_SORTING)
selected_rows: List[int] = ContextSetting([], schema_only=True)
sel_method: int = ContextSetting(SelectionMethods.N_BEST)
n_selected: int = ContextSetting(3)
class Inputs:
corpus = Input("Corpus", Corpus)
words = Input("Words", Table)
class Outputs:
selected_documents = Output("Selected documents", Corpus, default=True)
corpus = Output("Corpus", Corpus)
class Warning(OWWidget.Warning):
corpus_not_normalized = Msg("Use Preprocess Text to normalize corpus.")
class Error(OWWidget.Error):
custom_err = Msg("{}")
def __init__(self):
OWWidget.__init__(self)
ConcurrentWidgetMixin.__init__(self)
self._setup_control_area()
self._setup_main_area()
self.corpus = None
self.words = None
# saves scores avoid multiple computation of the same score
self.scores = {}
def _setup_control_area(self) -> None:
box = gui.widgetBox(self.controlArea, "Word Scoring Methods")
for value, (n, _, tt) in SCORING_METHODS.items():
b = gui.hBox(box, margin=0)
gui.checkBox(
b,
self,
value,
label=n,
callback=self.__setting_changed,
tooltip=tt,
)
if value in ADDITIONAL_OPTIONS:
value, options = ADDITIONAL_OPTIONS[value]
gui.comboBox(
b,
self,
value,
items=options,
callback=self.__setting_changed,
)
box = gui.widgetBox(self.controlArea, "Aggregation")
gui.comboBox(
box,
self,
"aggregation",
items=[n for n in AGGREGATIONS],
callback=self.__setting_changed,
)
gui.rubber(self.controlArea)
# select words box
box = gui.vBox(self.buttonsArea, "Select Documents")
grid = QGridLayout()
grid.setContentsMargins(0, 0, 0, 0)
self._sel_method_buttons = QButtonGroup()
for method, label in enumerate(SelectionMethods.ITEMS):
button = QRadioButton(label)
button.setChecked(method == self.sel_method)
grid.addWidget(button, method, 0)
self._sel_method_buttons.addButton(button, method)
self._sel_method_buttons.buttonClicked[int].connect(
self.__set_selection_method)
spin = gui.spin(
box,
self,
"n_selected",
1,
999,
addToLayout=False,
callback=lambda: self.__set_selection_method(SelectionMethods.
N_BEST),
)
grid.addWidget(spin, 3, 1)
box.layout().addLayout(grid)
# autocommit
gui.auto_send(self.buttonsArea, self, "auto_commit")
def _setup_main_area(self) -> None:
self._filter_line_edit = QLineEdit(
textChanged=self.__on_filter_changed, placeholderText="Filter...")
self.mainArea.layout().addWidget(self._filter_line_edit)
self.model = model = ScoreDocumentsTableModel(parent=self)
model.setHorizontalHeaderLabels(["Document"])
def select_manual():
self.__set_selection_method(SelectionMethods.MANUAL)
self.view = view = ScoreDocumentsTableView()
view.pressedAny.connect(select_manual)
self.mainArea.layout().addWidget(view)
# by default data are sorted in the Table order
header = self.view.horizontalHeader()
header.sectionClicked.connect(self.__on_horizontal_header_clicked)
proxy_model = ScoreDocumentsProxyModel()
proxy_model.setFilterKeyColumn(0)
proxy_model.setFilterCaseSensitivity(False)
view.setModel(proxy_model)
view.model().setSourceModel(self.model)
self.view.selectionModel().selectionChanged.connect(
self.__on_selection_change)
def __on_filter_changed(self) -> None:
model = self.view.model()
model.setFilterFixedString(self._filter_line_edit.text().strip())
def __on_horizontal_header_clicked(self, index: int):
header = self.view.horizontalHeader()
self.sort_column_order = (index, header.sortIndicatorOrder())
self._select_rows()
# when sorting change output table must consider the new order
# call explicitly since selection in table is not changed
if (self.sel_method == SelectionMethods.MANUAL and self.selected_rows
or self.sel_method == SelectionMethods.ALL):
# retrieve selection in new order
self.selected_rows = self.get_selected_indices()
self._send_output()
def __on_selection_change(self):
self.selected_rows = self.get_selected_indices()
self._send_output()
def __set_selection_method(self, method: int):
self.sel_method = method
self._sel_method_buttons.button(method).setChecked(True)
self._select_rows()
@Inputs.corpus
def set_data(self, corpus: Corpus) -> None:
self.closeContext()
self.Warning.corpus_not_normalized.clear()
if corpus is None:
self.corpus = None
self._clear_and_run()
return
if not self._is_corpus_normalized(corpus):
self.Warning.corpus_not_normalized()
self.corpus = corpus
self.selected_rows = []
self.openContext(corpus)
self._sel_method_buttons.button(self.sel_method).setChecked(True)
self._clear_and_run()
@staticmethod
def _get_word_attribute(words: Table) -> None:
attrs = [
a for a in words.domain.metas + words.domain.variables
if isinstance(a, StringVariable)
]
if not attrs:
return None
words_attr = next(
(a for a in attrs if a.attributes.get("type", "") == "words"),
None)
if words_attr:
return words.get_column_view(words_attr)[0].tolist()
else:
# find the most suitable attribute - one with lowest average text
# length - counted as a number of words
def avg_len(attr):
array_ = words.get_column_view(attr)[0]
array_ = array_[~isnull(array_)]
return sum(len(a.split()) for a in array_) / len(array_)
attr = sorted(attrs, key=avg_len)[0]
return words.get_column_view(attr)[0].tolist()
@Inputs.words
def set_words(self, words: Table) -> None:
if words is None or len(words.domain.variables +
words.domain.metas) == 0:
self.words = None
else:
self.words = self._get_word_attribute(words)
self._clear_and_run()
def _gather_scores(self) -> Tuple[np.ndarray, List[str]]:
"""
Gather scores and labels for the dictionary that holds scores
Returns
-------
scores
Scores table
labels
The list with score names for the header and variables names
"""
if self.corpus is None:
return np.empty((0, 0)), []
aggregation = self._get_active_aggregation()
scorers = self._get_active_scorers()
methods = [m for m in scorers if (m, aggregation) in self.scores]
scores = [self.scores[(m, aggregation)] for m in methods]
scores = np.column_stack(scores) if scores else np.empty(
(len(self.corpus), 0))
labels = [SCORING_METHODS[m][0] for m in methods]
return scores, labels
def _send_output(self) -> None:
"""
Create corpus with scores and output it
"""
if self.corpus is None:
self.Outputs.corpus.send(None)
self.Outputs.selected_documents.send(None)
return
scores, labels = self._gather_scores()
if labels:
d = self.corpus.domain
domain = Domain(
d.attributes,
d.class_var,
metas=d.metas + tuple(
ContinuousVariable(get_unique_names(d, l))
for l in labels),
)
out_corpus = Corpus(
domain,
self.corpus.X,
self.corpus.Y,
np.hstack([self.corpus.metas, scores]),
)
Corpus.retain_preprocessing(self.corpus, out_corpus)
else:
out_corpus = self.corpus
self.Outputs.corpus.send(
create_annotated_table(out_corpus, self.selected_rows))
self.Outputs.selected_documents.send(
out_corpus[self.selected_rows] if self.selected_rows else None)
def _fill_table(self) -> None:
"""
Fill the table in the widget with scores and document names
"""
if self.corpus is None:
self.model.clear()
return
scores, labels = self._gather_scores()
labels = ["Document"] + labels
titles = self.corpus.titles.tolist()
# clearing selection and sorting to prevent SEGFAULT on model.wrap
self.view.horizontalHeader().setSortIndicator(-1, Qt.AscendingOrder)
with disconnected(self.view.selectionModel().selectionChanged,
self.__on_selection_change):
self.view.clearSelection()
self.model.fill_table(titles, scores)
self.model.setHorizontalHeaderLabels(labels)
self.view.update_column_widths()
if self.model.columnCount() > self.sort_column_order[0]:
# if not enough columns do not apply sorting from settings since
# sorting can besaved for score column while scores are still computing
# tables is filled before scores are computed with document names
self.view.horizontalHeader().setSortIndicator(
*self.sort_column_order)
self._select_rows()
def _fill_and_output(self) -> None:
"""Fill the table in the widget and send the output"""
self._fill_table()
self._send_output()
def _clear_and_run(self) -> None:
"""Clear cached scores and commit"""
self.scores = {}
self.cancel()
self._fill_and_output()
self.commit()
def __setting_changed(self) -> None:
self.commit()
def commit(self) -> None:
self.Error.custom_err.clear()
self.cancel()
if self.corpus is not None and self.words is not None:
scorers = self._get_active_scorers()
aggregation = self._get_active_aggregation()
new_scores = [
s for s in scorers if (s, aggregation) not in self.scores
]
if new_scores:
self.start(
_run,
self.corpus,
self.words,
new_scores,
aggregation,
{
v: items[getattr(self, v)]
for v, items in ADDITIONAL_OPTIONS.values()
},
)
else:
self._fill_and_output()
def on_done(self, _: None) -> None:
self._send_output()
def on_partial_result(self, result: Tuple[str, str, np.ndarray]) -> None:
sc_method, aggregation, scores = result
self.scores[(sc_method, aggregation)] = scores
self._fill_table()
def on_exception(self, ex: Exception) -> None:
self.Error.custom_err(ex)
self._fill_and_output()
def _get_active_scorers(self) -> List[str]:
"""
Gather currently active/selected scores
Returns
-------
List with selected scores names
"""
return [attr for attr in SCORING_METHODS if getattr(self, attr)]
def _get_active_aggregation(self) -> str:
"""
Gather currently active/selected aggregation
Returns
-------
Selected aggregation name
"""
return list(AGGREGATIONS.keys())[self.aggregation]
@staticmethod
def _is_corpus_normalized(corpus: Corpus) -> bool:
"""
Check if corpus is normalized.
"""
return any(
isinstance(pp, BaseNormalizer)
for pp in corpus.used_preprocessor.preprocessors)
def get_selected_indices(self) -> List[int]:
# get indices in table's order - that the selected output table have same order
selected_rows = sorted(self.view.selectionModel().selectedRows(),
key=lambda idx: idx.row())
return [self.view.model().mapToSource(r).row() for r in selected_rows]
def _select_rows(self):
proxy_model = self.view.model()
n_rows, n_columns = proxy_model.rowCount(), proxy_model.columnCount()
if self.sel_method == SelectionMethods.NONE:
selection = QItemSelection()
elif self.sel_method == SelectionMethods.ALL:
selection = QItemSelection(
proxy_model.index(0, 0),
proxy_model.index(n_rows - 1, n_columns - 1))
elif self.sel_method == SelectionMethods.MANUAL:
selection = QItemSelection()
new_sel = []
for row in self.selected_rows:
if row < n_rows:
new_sel.append(row)
_selection = QItemSelection(
self.model.index(row, 0),
self.model.index(row, n_columns - 1))
selection.merge(
proxy_model.mapSelectionFromSource(_selection),
QItemSelectionModel.Select,
)
# selected rows must be updated when the same dataset with less rows
# appear at the input - it is not handled by selectionChanged
# in cases when all selected rows missing in new table
self.selected_rows = new_sel
elif self.sel_method == SelectionMethods.N_BEST:
n_sel = min(self.n_selected, n_rows)
selection = QItemSelection(
proxy_model.index(0, 0),
proxy_model.index(n_sel - 1, n_columns - 1))
else:
raise NotImplementedError
self.view.selectionModel().select(selection,
QItemSelectionModel.ClearAndSelect)
class SelectGenesEditor(ScBaseEditor):
DEFAULT_N_GENS = 1000
DEFAULT_METHOD = SelectMostVariableGenes.Dispersion
DEFAULT_COMPUTE_STATS = True
DEFAULT_N_GROUPS = 20
def __init__(self, parent=None, **kwargs):
super().__init__(parent, **kwargs)
self.setLayout(QVBoxLayout())
self._n_genes = self.DEFAULT_N_GENS
self._n_groups = self.DEFAULT_N_GROUPS
form = QFormLayout()
self.n_genes_spin = QSpinBox(minimum=1,
maximum=10**6,
value=self._n_genes)
self.n_genes_spin.valueChanged[int].connect(self._set_n_genes)
self.n_genes_spin.editingFinished.connect(self.edited)
form.addRow("Number of genes:", self.n_genes_spin)
self.layout().addLayout(form)
disp_b = QRadioButton("Dispersion", checked=True)
vari_b = QRadioButton("Variance")
mean_b = QRadioButton("Mean")
self.group = QButtonGroup()
self.group.buttonClicked.connect(self._on_button_clicked)
for i, button in enumerate([disp_b, vari_b, mean_b]):
index = index_to_enum(SelectMostVariableGenes.Method, i).value
self.group.addButton(button, index - 1)
form.addRow(button)
self.stats_check = QCheckBox("Compute statistics for",
checked=self.DEFAULT_COMPUTE_STATS)
self.stats_check.clicked.connect(self.edited)
self.n_groups_spin = QSpinBox(minimum=1, value=self._n_groups)
self.n_groups_spin.valueChanged[int].connect(self._set_n_groups)
self.n_groups_spin.editingFinished.connect(self.edited)
box = QHBoxLayout()
box.addWidget(self.stats_check)
box.addWidget(self.n_groups_spin)
box.addWidget(QLabel("gene groups."))
box.addStretch()
self.layout().addLayout(box)
def _set_n_genes(self, n):
if self._n_genes != n:
self._n_genes = n
self.n_genes_spin.setValue(n)
self.changed.emit()
def _set_n_groups(self, n):
if self._n_groups != n:
self._n_groups = n
self.n_groups_spin.setValue(n)
self.changed.emit()
def _on_button_clicked(self):
self.changed.emit()
self.edited.emit()
def setParameters(self, params):
self._set_n_genes(params.get("n_genes", self.DEFAULT_N_GENS))
method = params.get("method", self.DEFAULT_METHOD)
index = enum_to_index(SelectMostVariableGenes.Method, method)
self.group.buttons()[index].setChecked(True)
compute_stats = params.get("compute_stats", self.DEFAULT_COMPUTE_STATS)
self.stats_check.setChecked(compute_stats)
self._set_n_groups(params.get("n_groups", self.DEFAULT_N_GROUPS))
def parameters(self):
method = index_to_enum(SelectMostVariableGenes.Method,
self.group.checkedId())
return {
"n_genes": self._n_genes,
"method": method,
"compute_stats": self.stats_check.isChecked(),
"n_groups": self._n_groups
}
@staticmethod
def createinstance(params):
method = params.get("method", SelectGenesEditor.DEFAULT_METHOD)
n_genes = params.get("n_genes", SelectGenesEditor.DEFAULT_N_GENS)
compute_stats = params.get("compute_stats",
SelectGenesEditor.DEFAULT_COMPUTE_STATS)
n_groups = params.get("n_groups", SelectGenesEditor.DEFAULT_N_GROUPS) \
if compute_stats else None
return SelectMostVariableGenes(method, n_genes, n_groups)
def __repr__(self):
method = self.group.button(self.group.checkedId()).text()
text = "Method: {}, Number of Genes: {}".format(method, self._n_genes)
if self.stats_check.isChecked():
text += ", Number of Groups: {}".format(self._n_groups)
return text
class NormalizeEditor(ScBaseEditor):
DEFAULT_GROUP_BY = False
DEFAULT_GROUP_VAR = None
DEFAULT_METHOD = Normalize.CPM
def __init__(self, parent=None, master=None, **kwargs):
super().__init__(parent, **kwargs)
self._group_var = self.DEFAULT_GROUP_VAR
self._master = master
self._master.input_data_changed.connect(self._set_model)
self.setLayout(QVBoxLayout())
form = QFormLayout()
cpm_b = QRadioButton("Counts per million", checked=True)
med_b = QRadioButton("Median")
self.group = QButtonGroup()
self.group.buttonClicked.connect(self._on_button_clicked)
for i, button in enumerate([cpm_b, med_b]):
index = index_to_enum(Normalize.Method, i).value
self.group.addButton(button, index - 1)
form.addRow(button)
self.group_by_check = QCheckBox("Cell Groups: ",
enabled=self.DEFAULT_GROUP_BY)
self.group_by_check.clicked.connect(self.edited)
self.group_by_combo = QComboBox(enabled=self.DEFAULT_GROUP_BY)
self.group_by_model = DomainModel(order=(DomainModel.METAS,
DomainModel.CLASSES),
valid_types=DiscreteVariable,
alphabetical=True)
self.group_by_combo.setModel(self.group_by_model)
self.group_by_combo.currentIndexChanged.connect(self.changed)
self.group_by_combo.activated.connect(self.edited)
form.addRow(self.group_by_check, self.group_by_combo)
self.layout().addLayout(form)
self._set_model()
def _set_model(self):
data = self._master.data
self.group_by_model.set_domain(data and data.domain)
enable = bool(self.group_by_model)
self.group_by_check.setChecked(False)
self.group_by_check.setEnabled(enable)
self.group_by_combo.setEnabled(enable)
if self.group_by_model:
self.group_by_combo.setCurrentIndex(0)
if self._group_var and self._group_var in data.domain:
index = self.group_by_model.indexOf(self._group_var)
self.group_by_combo.setCurrentIndex(index)
else:
self.group_by_combo.setCurrentText(None)
def _on_button_clicked(self):
self.changed.emit()
self.edited.emit()
def setParameters(self, params):
method = params.get("method", self.DEFAULT_METHOD)
index = enum_to_index(Normalize.Method, method)
self.group.buttons()[index].setChecked(True)
self._group_var = params.get("group_var", self.DEFAULT_GROUP_VAR)
group = bool(self._group_var and self.group_by_model)
if group:
index = self.group_by_model.indexOf(self._group_var)
self.group_by_combo.setCurrentIndex(index)
group_by = params.get("group_by", self.DEFAULT_GROUP_BY)
self.group_by_check.setChecked(group_by and group)
def parameters(self):
index = self.group_by_combo.currentIndex()
group_var = self.group_by_model[index] if index > -1 else None
group_by = self.group_by_check.isChecked()
method = index_to_enum(Normalize.Method, self.group.checkedId())
return {"group_var": group_var, "group_by": group_by, "method": method}
@staticmethod
def createinstance(params):
group_var = params.get("group_var")
group_by = params.get("group_by", NormalizeEditor.DEFAULT_GROUP_BY)
method = params.get("method", NormalizeEditor.DEFAULT_METHOD)
return NormalizeGroups(group_var, method) \
if group_by and group_var else NormalizeSamples(method)
def __repr__(self):
method = self.group.button(self.group.checkedId()).text()
index = self.group_by_combo.currentIndex()
group_var = self.group_by_model[index] if index > -1 else None
group_by = self.group_by_check.isChecked()
group_text = ", Grouped by: {}".format(group_var) if group_by else ""
return "Method: {}".format(method) + group_text
class OWRank(OWWidget, ConcurrentWidgetMixin):
name = "排名(Rank)"
description = "根据数据特征的相关性对其进行排名和筛选。"
icon = "icons/Rank.svg"
priority = 1102
keywords = ['paiming', 'mingci', 'paixu']
category = "数据(Data)"
buttons_area_orientation = Qt.Vertical
class Inputs:
data = Input("数据(Data)", Table, replaces=['Data'])
scorer = MultiInput("评分器(Scorer)",
score.Scorer,
filter_none=True,
replaces=['Scorer'])
class Outputs:
reduced_data = Output("选中的数据(Reduced Data)",
Table,
default=True,
replaces=['Reduced Data'])
scores = Output("评分(Scores)", Table, replaces=['Scores'])
features = Output("特征(Features)",
AttributeList,
dynamic=False,
replaces=['Features'])
SelectNone, SelectAll, SelectManual, SelectNBest = range(4)
nSelected = ContextSetting(5)
auto_apply = Setting(True)
sorting = Setting((0, Qt.DescendingOrder))
selected_methods = Setting(set())
settings_version = 3
settingsHandler = DomainContextHandler()
selected_attrs = ContextSetting([], schema_only=True)
selectionMethod = ContextSetting(SelectNBest)
class Information(OWWidget.Information):
no_target_var = Msg("Data does not have a (single) target variable.")
missings_imputed = Msg('Missing values will be imputed as needed.')
class Error(OWWidget.Error):
invalid_type = Msg("Cannot handle target variable type {}")
inadequate_learner = Msg("Scorer {} inadequate: {}")
no_attributes = Msg("Data does not have a single attribute.")
class Warning(OWWidget.Warning):
renamed_variables = Msg(
"Variables with duplicated names have been renamed.")
def __init__(self):
OWWidget.__init__(self)
ConcurrentWidgetMixin.__init__(self)
self.scorers: List[ScoreMeta] = []
self.out_domain_desc = None
self.data = None
self.problem_type_mode = ProblemType.CLASSIFICATION
# results caches
self.scorers_results = {}
self.methods_results = {}
if not self.selected_methods:
self.selected_methods = {
method.name
for method in SCORES if method.is_default
}
# GUI
self.ranksModel = model = TableModel(parent=self) # type: TableModel
self.ranksView = view = TableView(self) # type: TableView
self.mainArea.layout().addWidget(view)
view.setModel(model)
view.setColumnWidth(NVAL_COL, 30)
view.selectionModel().selectionChanged.connect(self.on_select)
def _set_select_manual():
self.setSelectionMethod(OWRank.SelectManual)
view.manualSelection.connect(_set_select_manual)
view.verticalHeader().sectionClicked.connect(_set_select_manual)
view.horizontalHeader().sectionClicked.connect(self.headerClick)
self.measuresStack = stacked = QStackedWidget(self)
self.controlArea.layout().addWidget(stacked)
for scoring_methods in (CLS_SCORES, REG_SCORES, []):
box = gui.vBox(None, "评分方法" if scoring_methods else None)
stacked.addWidget(box)
for method in scoring_methods:
box.layout().addWidget(
QCheckBox(
method.zh_name,
self,
objectName=method.
shortname, # To be easily found in tests
checked=method.name in self.selected_methods,
stateChanged=partial(self.methodSelectionChanged,
method_name=method.name)))
gui.rubber(box)
gui.rubber(self.controlArea)
self.switchProblemType(ProblemType.CLASSIFICATION)
selMethBox = gui.vBox(self.buttonsArea, "选择特征")
grid = QGridLayout()
grid.setContentsMargins(0, 0, 0, 0)
grid.setSpacing(6)
self.selectButtons = QButtonGroup()
self.selectButtons.buttonClicked[int].connect(self.setSelectionMethod)
def button(text, buttonid, toolTip=None):
b = QRadioButton(text)
self.selectButtons.addButton(b, buttonid)
if toolTip is not None:
b.setToolTip(toolTip)
return b
b1 = button(self.tr("无"), OWRank.SelectNone)
b2 = button(self.tr("所有"), OWRank.SelectAll)
b3 = button(self.tr("手动"), OWRank.SelectManual)
b4 = button(self.tr("最佳排名:"), OWRank.SelectNBest)
s = gui.spin(
selMethBox,
self,
"nSelected",
1,
999,
callback=lambda: self.setSelectionMethod(OWRank.SelectNBest),
addToLayout=False)
grid.addWidget(b1, 0, 0)
grid.addWidget(b2, 1, 0)
grid.addWidget(b3, 2, 0)
grid.addWidget(b4, 3, 0)
grid.addWidget(s, 3, 1)
self.selectButtons.button(self.selectionMethod).setChecked(True)
selMethBox.layout().addLayout(grid)
gui.auto_send(self.buttonsArea, self, "auto_apply")
self.resize(690, 500)
def switchProblemType(self, index):
"""
Switch between discrete/continuous/no_class mode
"""
self.measuresStack.setCurrentIndex(index)
self.problem_type_mode = index
@Inputs.data
@check_sql_input
def set_data(self, data):
self.closeContext()
self.selected_attrs = []
self.ranksModel.clear()
self.ranksModel.resetSorting(True)
self.scorers_results = {}
self.methods_results = {}
self.cancel()
self.Error.clear()
self.Information.clear()
self.Information.missings_imputed(
shown=data is not None and data.has_missing())
if data is not None and not data.domain.attributes:
data = None
self.Error.no_attributes()
self.data = data
self.switchProblemType(ProblemType.CLASSIFICATION)
if self.data is not None:
domain = self.data.domain
if domain.has_discrete_class:
problem_type = ProblemType.CLASSIFICATION
elif domain.has_continuous_class:
problem_type = ProblemType.REGRESSION
elif not domain.class_var:
self.Information.no_target_var()
problem_type = ProblemType.UNSUPERVISED
else:
# This can happen?
self.Error.invalid_type(type(domain.class_var).__name__)
problem_type = None
if problem_type is not None:
self.switchProblemType(problem_type)
self.selectionMethod = OWRank.SelectNBest
self.openContext(data)
self.selectButtons.button(self.selectionMethod).setChecked(True)
def handleNewSignals(self):
self.setStatusMessage('Running')
self.update_scores()
self.setStatusMessage('')
self.on_select()
@Inputs.scorer
def set_learner(self, index, scorer):
self.scorers[index] = ScoreMeta(
scorer.name, scorer.name, scorer,
ProblemType.from_variable(scorer.class_type), False)
self.scorers_results = {}
@Inputs.scorer.insert
def insert_learner(self, index: int, scorer):
self.scorers.insert(
index,
ScoreMeta(scorer.name, scorer.name, scorer,
ProblemType.from_variable(scorer.class_type), False))
self.scorers_results = {}
@Inputs.scorer.remove
def remove_learner(self, index):
self.scorers.pop(index)
self.scorers_results = {}
def _get_methods(self):
return [
method for method in SCORES if
(method.name in self.selected_methods
and method.problem_type == self.problem_type_mode and
(not issparse(self.data.X) or method.scorer.supports_sparse_data))
]
def _get_scorers(self):
scorers = []
for scorer in self.scorers:
if scorer.problem_type in (
self.problem_type_mode,
ProblemType.UNSUPERVISED,
):
scorers.append(scorer)
else:
self.Error.inadequate_learner(scorer.name,
scorer.learner_adequacy_err_msg)
return scorers
def update_scores(self):
if self.data is None:
self.ranksModel.clear()
self.Outputs.scores.send(None)
return
self.Error.inadequate_learner.clear()
scorers = [
s for s in self._get_scorers() if s not in self.scorers_results
]
methods = [
m for m in self._get_methods() if m not in self.methods_results
]
self.start(run, self.data, methods, scorers)
def on_done(self, result: Results) -> None:
self.methods_results.update(result.method_scores)
self.scorers_results.update(result.scorer_scores)
methods = self._get_methods()
method_labels = tuple(m.zh_shortname for m in methods)
method_scores = tuple(self.methods_results[m] for m in methods)
scores = [self.scorers_results[s] for s in self._get_scorers()]
scorer_scores, scorer_labels = zip(*scores) if scores else ((), ())
labels = method_labels + tuple(chain.from_iterable(scorer_labels))
model_array = np.column_stack(
(list(self.data.domain.attributes), ) + ([
float(len(a.values)) if a.is_discrete else np.nan
for a in self.data.domain.attributes
], ) + method_scores + scorer_scores)
for column, values in enumerate(model_array.T[2:].astype(float),
start=2):
self.ranksModel.setExtremesFrom(column, values)
self.ranksModel.wrap(model_array.tolist())
self.ranksModel.setHorizontalHeaderLabels((
'',
'#',
) + labels)
self.ranksView.setColumnWidth(NVAL_COL, 40)
self.ranksView.resizeColumnToContents(VARNAME_COL)
# Re-apply sort
try:
sort_column, sort_order = self.sorting
if sort_column < len(labels):
# adds 2 to skip the first two columns
self.ranksModel.sort(sort_column + 2, sort_order)
self.ranksView.horizontalHeader().setSortIndicator(
sort_column + 2, sort_order)
except ValueError:
pass
self.autoSelection()
self.Outputs.scores.send(self.create_scores_table(labels))
def on_exception(self, ex: Exception) -> None:
raise ex
def on_partial_result(self, result: Any) -> None:
pass
def on_select(self):
# Save indices of attributes in the original, unsorted domain
selected_rows = self.ranksView.selectionModel().selectedRows(0)
row_indices = [i.row() for i in selected_rows]
attr_indices = self.ranksModel.mapToSourceRows(row_indices)
self.selected_attrs = [self.data.domain[idx] for idx in attr_indices]
self.commit.deferred()
def setSelectionMethod(self, method):
self.selectionMethod = method
self.selectButtons.button(method).setChecked(True)
self.autoSelection()
def autoSelection(self):
selModel = self.ranksView.selectionModel()
model = self.ranksModel
rowCount = model.rowCount()
columnCount = model.columnCount()
if self.selectionMethod == OWRank.SelectNone:
selection = QItemSelection()
elif self.selectionMethod == OWRank.SelectAll:
selection = QItemSelection(
model.index(0, 0), model.index(rowCount - 1, columnCount - 1))
elif self.selectionMethod == OWRank.SelectNBest:
nSelected = min(self.nSelected, rowCount)
selection = QItemSelection(
model.index(0, 0), model.index(nSelected - 1, columnCount - 1))
else:
selection = QItemSelection()
if self.selected_attrs is not None:
attr_indices = [
self.data.domain.attributes.index(var)
for var in self.selected_attrs
]
for row in model.mapFromSourceRows(attr_indices):
selection.append(
QItemSelectionRange(model.index(row, 0),
model.index(row, columnCount - 1)))
selModel.select(selection, QItemSelectionModel.ClearAndSelect)
def headerClick(self, index):
if index >= 2 and self.selectionMethod == OWRank.SelectNBest:
# Reselect the top ranked attributes
self.autoSelection()
# Store the header states
sort_order = self.ranksModel.sortOrder()
# -2 for '#' (discrete count) column
sort_column = self.ranksModel.sortColumn() - 2
self.sorting = (sort_column, sort_order)
def methodSelectionChanged(self, state, method_name):
if state == Qt.Checked:
self.selected_methods.add(method_name)
elif method_name in self.selected_methods:
self.selected_methods.remove(method_name)
self.update_scores()
def send_report(self):
if not self.data:
return
self.report_domain("Input", self.data.domain)
self.report_table("Ranks", self.ranksView, num_format="{:.3f}")
if self.out_domain_desc is not None:
self.report_items("Output", self.out_domain_desc)
@gui.deferred
def commit(self):
if not self.selected_attrs:
self.Outputs.reduced_data.send(None)
self.Outputs.features.send(None)
self.out_domain_desc = None
else:
reduced_domain = Domain(self.selected_attrs,
self.data.domain.class_var,
self.data.domain.metas)
data = self.data.transform(reduced_domain)
self.Outputs.reduced_data.send(data)
self.Outputs.features.send(AttributeList(self.selected_attrs))
self.out_domain_desc = report.describe_domain(data.domain)
def create_scores_table(self, labels):
self.Warning.renamed_variables.clear()
model_list = self.ranksModel.tolist()
# Empty or just first two columns
if not model_list or len(model_list[0]) == 2:
return None
unique, renamed = get_unique_names_duplicates(labels + ('Feature', ),
return_duplicated=True)
if renamed:
self.Warning.renamed_variables(', '.join(renamed))
domain = Domain([ContinuousVariable(label) for label in unique[:-1]],
metas=[StringVariable(unique[-1])])
# Prevent np.inf scores
finfo = np.finfo(np.float64)
scores = np.clip(np.array(model_list)[:, 2:], finfo.min, finfo.max)
feature_names = np.array([a.name for a in self.data.domain.attributes])
# Reshape to 2d array as Table does not like 1d arrays
feature_names = feature_names[:, None]
new_table = Table(domain, scores, metas=feature_names)
new_table.name = "Feature Scores"
return new_table
@classmethod
def migrate_settings(cls, settings, version):
# If older settings, restore sort header to default
# Saved selected_rows will likely be incorrect
if version is None or version < 2:
column, order = 0, Qt.DescendingOrder
headerState = settings.pop("headerState", None)
# Lacking knowledge of last problemType, use discrete ranks view's ordering
if isinstance(headerState, (tuple, list)):
headerState = headerState[0]
if isinstance(headerState, bytes):
hview = QHeaderView(Qt.Horizontal)
hview.restoreState(headerState)
column, order = hview.sortIndicatorSection(
) - 1, hview.sortIndicatorOrder()
settings["sorting"] = (column, order)
@classmethod
def migrate_context(cls, context, version):
if version is None or version < 3:
# Selections were stored as indices, so these contexts matched
# any domain. The only safe thing to do is to remove them.
raise IncompatibleContext
class OWRankSurvivalFeatures(OWWidget, ConcurrentWidgetMixin):
name = 'Rank Survival Features'
# TODO: Add widget metadata
description = ''
icon = 'icons/owranksurvivalfeatures.svg'
priority = 30
keywords = []
buttons_area_orientation = Qt.Vertical
select_none, manual_selection, select_n_best = range(3)
settingsHandler = DomainContextHandler()
selected_attrs = ContextSetting([], schema_only=True)
selection_method = Setting(select_n_best, schema_only=True)
n_selected = Setting(20, schema_only=True)
auto_commit: bool = Setting(False, schema_only=True)
class Inputs:
data = Input('Data', Table)
class Outputs:
reduced_data = Output('Reduced Data', Table, default=True)
def __init__(self):
OWWidget.__init__(self)
ConcurrentWidgetMixin.__init__(self)
self.data: Optional[Table] = None
self.attr_name_to_variable: Optional[Table] = None
self.covariates_from_worker_result = None
self.time_var: Optional[str] = None
self.event_var: Optional[str] = None
gui.rubber(self.controlArea)
sel_method_box = gui.vBox(self.buttonsArea, 'Select Attributes')
grid = QGridLayout()
grid.setContentsMargins(0, 0, 0, 0)
grid.setSpacing(6)
self.select_buttons = QButtonGroup()
self.select_buttons.buttonClicked[int].connect(
self.set_selection_method)
def button(text, buttonid, toolTip=None):
b = QRadioButton(text)
self.select_buttons.addButton(b, buttonid)
if toolTip is not None:
b.setToolTip(toolTip)
return b
b1 = button(self.tr('None'), OWRankSurvivalFeatures.select_none)
b2 = button(self.tr('Manual'), OWRankSurvivalFeatures.manual_selection)
b3 = button(self.tr('Best ranked:'),
OWRankSurvivalFeatures.select_n_best)
s = gui.spin(
sel_method_box,
self,
'n_selected',
1,
999,
callback=lambda: self.set_selection_method(OWRankSurvivalFeatures.
select_n_best),
addToLayout=False,
)
grid.addWidget(b1, 0, 0)
grid.addWidget(b2, 1, 0)
grid.addWidget(b3, 2, 0)
grid.addWidget(s, 2, 1)
sel_method_box.layout().addLayout(grid)
self.commit_button = gui.auto_commit(self.buttonsArea,
self,
'auto_commit',
'&Commit',
box=False)
# Main area
self.model = PyTableModel(parent=self)
self.table_view = TableView(parent=self)
self.table_view.setModel(self.model)
self.model.setHorizontalHeaderLabels([
'Log-Likelihood', 'Log-Likelihood Ratio', f'{"p".center(13)}',
'FDR'
])
self.table_view.setSizeAdjustPolicy(
QAbstractScrollArea.AdjustToContentsOnFirstShow)
self.table_view.selectionModel().selectionChanged.connect(
self.on_select)
def _set_select_manual():
self.set_selection_method(OWRankSurvivalFeatures.manual_selection)
self.table_view.manualSelection.connect(_set_select_manual)
self.table_view.verticalHeader().sectionClicked.connect(
_set_select_manual)
self.mainArea.layout().addWidget(self.table_view)
@property
def covariates(self) -> Optional[List[str]]:
if not self.data:
return
return [attr.name for attr in self.data.domain.attributes]
@Inputs.data
@check_survival_data
def set_data(self, data: Table):
self.closeContext()
self.selected_attrs = []
self.covariates_from_worker_result = []
self.model.clear()
self.model.resetSorting()
if not data:
return
self.data = data
self.attr_name_to_variable = {
attr.name: attr
for attr in self.data.domain.attributes
}
self.openContext(data)
time_var, event_var = get_survival_endpoints(self.data.domain)
self.time_var, self.event_var = time_var.name, event_var.name
self.start(worker, self.data, self.covariates, self.time_var,
self.event_var)
def commit(self):
if not self.selected_attrs:
self.Outputs.reduced_data.send(None)
else:
reduced_domain = Domain(self.selected_attrs,
self.data.domain.class_vars,
self.data.domain.metas)
data = self.data.transform(reduced_domain)
self.Outputs.reduced_data.send(data)
def on_done(self, worker_result):
covariate_names, results = worker_result
# wrap everything except covariate names
self.model.wrap(results.tolist())
# this is temp solution because covariate orders gets mixed when using multiprocessing
self.covariates_from_worker_result = covariate_names.tolist()
# match covariate names to domain variables and set vertical header
self.model.setVerticalHeaderLabels(
[self.attr_name_to_variable[name] for name in covariate_names])
self.table_view.resizeColumnsToContents()
self.auto_select()
def on_exception(self, ex):
raise ex
def on_partial_result(self, result: Any) -> None:
pass
def set_selection_method(self, method):
self.selection_method = method
self.select_buttons.button(method).setChecked(True)
self.auto_select()
def auto_select(self):
selection_model = self.table_view.selectionModel()
row_count = self.model.rowCount()
column_count = self.model.columnCount()
if self.selection_method == OWRankSurvivalFeatures.select_none:
selection = QItemSelection()
elif self.selection_method == OWRankSurvivalFeatures.select_n_best:
n_selected = min(self.n_selected, row_count)
selection = QItemSelection(
self.model.index(0, 0),
self.model.index(n_selected - 1, column_count - 1))
else:
selection = QItemSelection()
if self.selected_attrs is not None:
attr_indices = [
self.covariates_from_worker_result.index(var.name)
for var in self.selected_attrs
]
for row in self.model.mapFromSourceRows(attr_indices):
selection.append(
QItemSelectionRange(
self.model.index(row, 0),
self.model.index(row, column_count - 1)))
selection_model.select(selection, QItemSelectionModel.ClearAndSelect)
def on_select(self):
selected_rows = self.table_view.selectionModel().selectedRows(0)
row_indices = [i.row() for i in selected_rows]
attr_indices = self.model.mapToSourceRows(row_indices)
self.selected_attrs = [
self.model._headers[Qt.Vertical][row] for row in attr_indices
]
self.commit()