def __init__(self, parent=None, signalManager=None):
OWWidget.__init__(self, parent, signalManager, "Nomogram", 1)
#self.setWFlags(Qt.WResizeNoErase | Qt.WRepaintNoErase) #this works like magic.. no flicker during repaint!
self.parent = parent
# self.setWFlags(self.getWFlags()+Qt.WStyle_Maximize)
self.callbackDeposit = [] # deposit for OWGUI callback functions
self.alignType = 0
self.contType = 0
self.yAxis = 0
self.probability = 0
self.verticalSpacing = 60
self.verticalSpacingContinuous = 100
self.diff_between_ordinal = 30
self.fontSize = 9
self.lineWidth = 1
self.histogram = 0
self.histogram_size = 10
self.data = None
self.cl = None
self.confidence_check = 0
self.confidence_percent = 95
self.sort_type = 0
self.loadSettings()
self.pointsName = ["Total", "Total"]
self.totalPointsName = ["Probability", "Probability"]
self.bnomogram = None
self.inputs = [("Classifier", orange.Classifier, self.classifier)]
self.TargetClassIndex = 0
self.targetCombo = OWGUI.comboBox(
self.controlArea,
self,
"TargetClassIndex",
" Target Class ",
addSpace=True,
tooltip='Select target (prediction) class in the model.',
callback=self.setTarget)
self.alignRadio = OWGUI.radioButtonsInBox(
self.controlArea,
self,
'alignType', ['Align left', 'Align by zero influence'],
box='Attribute placement',
tooltips=[
'Attributes in nomogram are left aligned',
'Attributes are not aligned, top scale represents true (normalized) regression coefficient value'
],
addSpace=True,
callback=self.showNomogram)
self.verticalSpacingLabel = OWGUI.spin(
self.alignRadio,
self,
'verticalSpacing',
15,
200,
label='Vertical spacing:',
orientation=0,
tooltip='Define space (pixels) between adjacent attributes.',
callback=self.showNomogram)
self.ContRadio = OWGUI.radioButtonsInBox(
self.controlArea,
self,
'contType', ['1D projection', '2D curve'],
'Continuous attributes',
tooltips=[
'Continuous attribute are presented on a single scale',
'Two dimensional space is used to present continuous attributes in nomogram.'
],
addSpace=True,
callback=[
lambda: self.verticalSpacingContLabel.setDisabled(
not self.contType), self.showNomogram
])
self.verticalSpacingContLabel = OWGUI.spin(
OWGUI.indentedBox(self.ContRadio,
sep=OWGUI.checkButtonOffsetHint(
self.ContRadio.buttons[-1])),
self,
'verticalSpacingContinuous',
15,
200,
label="Height",
orientation=0,
tooltip=
'Define space (pixels) between adjacent 2d presentation of attributes.',
callback=self.showNomogram)
self.verticalSpacingContLabel.setDisabled(not self.contType)
self.yAxisRadio = OWGUI.radioButtonsInBox(
self.controlArea,
self,
'yAxis', ['Point scale', 'Log odds ratios'],
'Scale',
tooltips=[
'values are normalized on a 0-100 point scale',
'values on top axis show log-linear contribution of attribute to full model'
],
addSpace=True,
callback=self.showNomogram)
layoutBox = OWGUI.widgetBox(self.controlArea,
"Display",
orientation=1,
addSpace=True)
self.probabilityCheck = OWGUI.checkBox(layoutBox,
self,
'probability',
'Show prediction',
tooltip='',
callback=self.setProbability)
self.CICheck, self.CILabel = OWGUI.checkWithSpin(
layoutBox,
self,
'Confidence intervals (%):',
min=1,
max=99,
step=1,
checked='confidence_check',
value='confidence_percent',
checkCallback=self.showNomogram,
spinCallback=self.showNomogram)
self.histogramCheck, self.histogramLabel = OWGUI.checkWithSpin(
layoutBox,
self,
'Show histogram, size',
min=1,
max=30,
checked='histogram',
value='histogram_size',
step=1,
tooltip='-(TODO)-',
checkCallback=self.showNomogram,
spinCallback=self.showNomogram)
OWGUI.separator(layoutBox)
self.sortOptions = [
"No sorting", "Absolute importance", "Positive influence",
"Negative influence"
]
self.sortBox = OWGUI.comboBox(layoutBox,
self,
"sort_type",
label="Sort by ",
items=self.sortOptions,
callback=self.sortNomogram,
orientation="horizontal")
OWGUI.rubber(self.controlArea)
self.connect(self.graphButton, SIGNAL("clicked()"),
self.menuItemPrinter)
#add a graph widget
self.header = OWNomogramHeader(None, self.mainArea)
self.header.setFixedHeight(60)
self.header.setVerticalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.header.setHorizontalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.graph = OWNomogramGraph(self.bnomogram, self.mainArea)
self.graph.setMinimumWidth(200)
self.graph.setHorizontalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.footer = OWNomogramHeader(None, self.mainArea)
self.footer.setFixedHeight(60 * 2 + 10)
self.footer.setVerticalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.footer.setHorizontalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.mainArea.layout().addWidget(self.header)
self.mainArea.layout().addWidget(self.graph)
self.mainArea.layout().addWidget(self.footer)
self.resize(700, 500)
#self.repaint()
#self.update()
# mouse pressed flag
self.mousepr = False
def __init__(self,parent=None, signalManager = None):
OWWidget.__init__(self, parent, signalManager, "Nomogram", 1)
#self.setWFlags(Qt.WResizeNoErase | Qt.WRepaintNoErase) #this works like magic.. no flicker during repaint!
self.parent = parent
# self.setWFlags(self.getWFlags()+Qt.WStyle_Maximize)
self.callbackDeposit = [] # deposit for OWGUI callback functions
self.alignType = 0
self.contType = 0
self.yAxis = 0
self.probability = 0
self.verticalSpacing = 60
self.verticalSpacingContinuous = 100
self.diff_between_ordinal = 30
self.fontSize = 9
self.lineWidth = 1
self.histogram = 0
self.histogram_size = 10
self.data = None
self.cl = None
self.confidence_check = 0
self.confidence_percent = 95
self.sort_type = 0
self.loadSettings()
self.pointsName = ["Total", "Total"]
self.totalPointsName = ["Probability", "Probability"]
self.bnomogram = None
self.inputs=[("Classifier", orange.Classifier, self.classifier)]
self.TargetClassIndex = 0
self.targetCombo = OWGUI.comboBox(self.controlArea, self, "TargetClassIndex", " Target Class ", addSpace=True, tooltip='Select target (prediction) class in the model.', callback = self.setTarget)
self.alignRadio = OWGUI.radioButtonsInBox(self.controlArea, self, 'alignType', ['Align left', 'Align by zero influence'], box='Attribute placement',
tooltips=['Attributes in nomogram are left aligned', 'Attributes are not aligned, top scale represents true (normalized) regression coefficient value'],
addSpace=True,
callback=self.showNomogram)
self.verticalSpacingLabel = OWGUI.spin(self.alignRadio, self, 'verticalSpacing', 15, 200, label = 'Vertical spacing:', orientation = 0, tooltip='Define space (pixels) between adjacent attributes.', callback = self.showNomogram)
self.ContRadio = OWGUI.radioButtonsInBox(self.controlArea, self, 'contType', ['1D projection', '2D curve'], 'Continuous attributes',
tooltips=['Continuous attribute are presented on a single scale', 'Two dimensional space is used to present continuous attributes in nomogram.'],
addSpace=True,
callback=[lambda:self.verticalSpacingContLabel.setDisabled(not self.contType), self.showNomogram])
self.verticalSpacingContLabel = OWGUI.spin(OWGUI.indentedBox(self.ContRadio, sep=OWGUI.checkButtonOffsetHint(self.ContRadio.buttons[-1])), self, 'verticalSpacingContinuous', 15, 200, label = "Height", orientation=0, tooltip='Define space (pixels) between adjacent 2d presentation of attributes.', callback = self.showNomogram)
self.verticalSpacingContLabel.setDisabled(not self.contType)
self.yAxisRadio = OWGUI.radioButtonsInBox(self.controlArea, self, 'yAxis', ['Point scale', 'Log odds ratios'], 'Scale',
tooltips=['values are normalized on a 0-100 point scale','values on top axis show log-linear contribution of attribute to full model'],
addSpace=True,
callback=self.showNomogram)
layoutBox = OWGUI.widgetBox(self.controlArea, "Display", orientation=1, addSpace=True)
self.probabilityCheck = OWGUI.checkBox(layoutBox, self, 'probability', 'Show prediction', tooltip='', callback = self.setProbability)
self.CICheck, self.CILabel = OWGUI.checkWithSpin(layoutBox, self, 'Confidence intervals (%):', min=1, max=99, step = 1, checked='confidence_check', value='confidence_percent', checkCallback=self.showNomogram, spinCallback = self.showNomogram)
self.histogramCheck, self.histogramLabel = OWGUI.checkWithSpin(layoutBox, self, 'Show histogram, size', min=1, max=30, checked='histogram', value='histogram_size', step = 1, tooltip='-(TODO)-', checkCallback=self.showNomogram, spinCallback = self.showNomogram)
OWGUI.separator(layoutBox)
self.sortOptions = ["No sorting", "Absolute importance", "Positive influence", "Negative influence"]
self.sortBox = OWGUI.comboBox(layoutBox, self, "sort_type", label="Sort by ", items=self.sortOptions, callback = self.sortNomogram, orientation="horizontal")
OWGUI.rubber(self.controlArea)
self.connect(self.graphButton, SIGNAL("clicked()"), self.menuItemPrinter)
#add a graph widget
self.header = OWNomogramHeader(None, self.mainArea)
self.header.setFixedHeight(60)
self.header.setVerticalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.header.setHorizontalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.graph = OWNomogramGraph(self.bnomogram, self.mainArea)
self.graph.setMinimumWidth(200)
self.graph.setHorizontalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.footer = OWNomogramHeader(None, self.mainArea)
self.footer.setFixedHeight(60*2+10)
self.footer.setVerticalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.footer.setHorizontalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.mainArea.layout().addWidget(self.header)
self.mainArea.layout().addWidget(self.graph)
self.mainArea.layout().addWidget(self.footer)
self.resize(700,500)
#self.repaint()
#self.update()
# mouse pressed flag
self.mousepr = False
class OWNomogram(OWWidget):
settingsList = [
"alignType", "verticalSpacing", "contType",
"verticalSpacingContinuous", "yAxis", "probability",
"confidence_check", "confidence_percent", "histogram",
"histogram_size", "sort_type"
]
contextHandlers = {
"": DomainContextHandler("", ["TargetClassIndex"], matchValues=1)
}
def __init__(self, parent=None, signalManager=None):
OWWidget.__init__(self, parent, signalManager, "Nomogram", 1)
#self.setWFlags(Qt.WResizeNoErase | Qt.WRepaintNoErase) #this works like magic.. no flicker during repaint!
self.parent = parent
# self.setWFlags(self.getWFlags()+Qt.WStyle_Maximize)
self.callbackDeposit = [] # deposit for OWGUI callback functions
self.alignType = 0
self.contType = 0
self.yAxis = 0
self.probability = 0
self.verticalSpacing = 60
self.verticalSpacingContinuous = 100
self.diff_between_ordinal = 30
self.fontSize = 9
self.lineWidth = 1
self.histogram = 0
self.histogram_size = 10
self.data = None
self.cl = None
self.confidence_check = 0
self.confidence_percent = 95
self.sort_type = 0
self.loadSettings()
self.pointsName = ["Total", "Total"]
self.totalPointsName = ["Probability", "Probability"]
self.bnomogram = None
self.inputs = [("Classifier", orange.Classifier, self.classifier)]
self.TargetClassIndex = 0
self.targetCombo = OWGUI.comboBox(
self.controlArea,
self,
"TargetClassIndex",
" Target Class ",
addSpace=True,
tooltip='Select target (prediction) class in the model.',
callback=self.setTarget)
self.alignRadio = OWGUI.radioButtonsInBox(
self.controlArea,
self,
'alignType', ['Align left', 'Align by zero influence'],
box='Attribute placement',
tooltips=[
'Attributes in nomogram are left aligned',
'Attributes are not aligned, top scale represents true (normalized) regression coefficient value'
],
addSpace=True,
callback=self.showNomogram)
self.verticalSpacingLabel = OWGUI.spin(
self.alignRadio,
self,
'verticalSpacing',
15,
200,
label='Vertical spacing:',
orientation=0,
tooltip='Define space (pixels) between adjacent attributes.',
callback=self.showNomogram)
self.ContRadio = OWGUI.radioButtonsInBox(
self.controlArea,
self,
'contType', ['1D projection', '2D curve'],
'Continuous attributes',
tooltips=[
'Continuous attribute are presented on a single scale',
'Two dimensional space is used to present continuous attributes in nomogram.'
],
addSpace=True,
callback=[
lambda: self.verticalSpacingContLabel.setDisabled(
not self.contType), self.showNomogram
])
self.verticalSpacingContLabel = OWGUI.spin(
OWGUI.indentedBox(self.ContRadio,
sep=OWGUI.checkButtonOffsetHint(
self.ContRadio.buttons[-1])),
self,
'verticalSpacingContinuous',
15,
200,
label="Height",
orientation=0,
tooltip=
'Define space (pixels) between adjacent 2d presentation of attributes.',
callback=self.showNomogram)
self.verticalSpacingContLabel.setDisabled(not self.contType)
self.yAxisRadio = OWGUI.radioButtonsInBox(
self.controlArea,
self,
'yAxis', ['Point scale', 'Log odds ratios'],
'Scale',
tooltips=[
'values are normalized on a 0-100 point scale',
'values on top axis show log-linear contribution of attribute to full model'
],
addSpace=True,
callback=self.showNomogram)
layoutBox = OWGUI.widgetBox(self.controlArea,
"Display",
orientation=1,
addSpace=True)
self.probabilityCheck = OWGUI.checkBox(layoutBox,
self,
'probability',
'Show prediction',
tooltip='',
callback=self.setProbability)
self.CICheck, self.CILabel = OWGUI.checkWithSpin(
layoutBox,
self,
'Confidence intervals (%):',
min=1,
max=99,
step=1,
checked='confidence_check',
value='confidence_percent',
checkCallback=self.showNomogram,
spinCallback=self.showNomogram)
self.histogramCheck, self.histogramLabel = OWGUI.checkWithSpin(
layoutBox,
self,
'Show histogram, size',
min=1,
max=30,
checked='histogram',
value='histogram_size',
step=1,
tooltip='-(TODO)-',
checkCallback=self.showNomogram,
spinCallback=self.showNomogram)
OWGUI.separator(layoutBox)
self.sortOptions = [
"No sorting", "Absolute importance", "Positive influence",
"Negative influence"
]
self.sortBox = OWGUI.comboBox(layoutBox,
self,
"sort_type",
label="Sort by ",
items=self.sortOptions,
callback=self.sortNomogram,
orientation="horizontal")
OWGUI.rubber(self.controlArea)
self.connect(self.graphButton, SIGNAL("clicked()"),
self.menuItemPrinter)
#add a graph widget
self.header = OWNomogramHeader(None, self.mainArea)
self.header.setFixedHeight(60)
self.header.setVerticalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.header.setHorizontalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.graph = OWNomogramGraph(self.bnomogram, self.mainArea)
self.graph.setMinimumWidth(200)
self.graph.setHorizontalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.footer = OWNomogramHeader(None, self.mainArea)
self.footer.setFixedHeight(60 * 2 + 10)
self.footer.setVerticalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.footer.setHorizontalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.mainArea.layout().addWidget(self.header)
self.mainArea.layout().addWidget(self.graph)
self.mainArea.layout().addWidget(self.footer)
self.resize(700, 500)
#self.repaint()
#self.update()
# mouse pressed flag
self.mousepr = False
def sendReport(self):
if self.cl:
tclass = self.cl.domain.classVar.values[self.TargetClassIndex]
else:
tclass = "N/A"
self.reportSettings(
"Information",
[("Target class", tclass), self.confidence_check and
("Confidence intervals", "%i%%" % self.confidence_percent),
("Sorting",
self.sortOptions[self.sort_type] if self.sort_type else "None")])
if self.cl:
canvases = header, graph, footer = self.header.scene(
), self.graph.scene(), self.footer.scene()
painter = QPainter()
buffer = QPixmap(max(c.width() for c in canvases),
sum(c.height() for c in canvases))
painter.begin(buffer)
painter.fillRect(buffer.rect(), QBrush(QColor(255, 255, 255)))
header.render(painter, QRectF(0, 0, header.width(),
header.height()),
QRectF(0, 0, header.width(), header.height()))
graph.render(
painter,
QRectF(0, header.height(), graph.width(), graph.height()),
QRectF(0, 0, graph.width(), graph.height()))
footer.render(
painter,
QRectF(0,
header.height() + graph.height(), footer.width(),
footer.height()),
QRectF(0, 0, footer.width(), footer.height()))
painter.end()
self.reportImage(lambda filename: buffer.save(
filename,
os.path.splitext(filename)[1][1:]))
# Input channel: the Bayesian classifier
def nbClassifier(self, cl):
# this subroutine computes standard error of estimated beta. Note that it is used only for discrete data,
# continuous data have a different computation.
def errOld(e, priorError, key, data):
inf = 0.0
sume = e[0] + e[1]
for d in data:
if d[at] == key:
inf += (e[0] * e[1] / sume / sume)
inf = max(inf, aproxZero)
var = max(1 / inf - priorError * priorError, 0)
return (math.sqrt(var))
def err(condDist, att, value, targetClass, priorError, data):
sumE = sum(condDist)
valueE = condDist[targetClass]
distAtt = orange.Distribution(att, data)
inf = distAtt[value] * (valueE / sumE) * (1 - valueE / sumE)
inf = max(inf, aproxZero)
var = max(1 / inf - priorError * priorError, 0)
return (math.sqrt(var))
classVal = cl.domain.classVar
att = cl.domain.attributes
# calculate prior probability
dist1 = max(aproxZero,
1 - cl.distribution[classVal[self.TargetClassIndex]])
dist0 = max(aproxZero,
cl.distribution[classVal[self.TargetClassIndex]])
prior = dist0 / dist1
if self.data:
sumd = dist1 + dist0
priorError = math.sqrt(
1 / ((dist1 * dist0 / sumd / sumd) * len(self.data)))
else:
priorError = 0
if self.bnomogram:
self.bnomogram.destroy_and_init(
self, AttValue("Constant", math.log(prior), error=priorError))
else:
self.bnomogram = BasicNomogram(
self, AttValue("Constant", math.log(prior), error=priorError))
if self.data:
stat = getCached(self.data, orange.DomainBasicAttrStat,
(self.data, ))
for at in range(len(att)):
a = None
if att[at].varType == orange.VarTypes.Discrete:
if att[at].ordered:
a = AttrLineOrdered(att[at].name, self.bnomogram)
else:
a = AttrLine(att[at].name, self.bnomogram)
for cd in cl.conditionalDistributions[at].keys():
# calculuate thickness
conditional0 = max(
cl.conditionalDistributions[at][cd][classVal[
self.TargetClassIndex]], aproxZero)
conditional1 = max(
1 - cl.conditionalDistributions[at][cd][classVal[
self.TargetClassIndex]], aproxZero)
beta = math.log(conditional0 / conditional1 / prior)
if self.data:
#thickness = int(round(4.*float(len(self.data.filter({att[at].name:str(cd)})))/float(len(self.data))))
thickness = float(
len(self.data.filter({att[at].name: str(cd)
}))) / float(len(self.data))
se = err(cl.conditionalDistributions[at][cd], att[at],
cd, classVal[self.TargetClassIndex],
priorError,
self.data) # standar error of beta
else:
thickness = 0
se = 0
a.addAttValue(
AttValue(str(cd), beta, lineWidth=thickness, error=se))
else:
a = AttrLineCont(att[at].name, self.bnomogram)
numOfPartitions = 50
if self.data:
maxAtValue = stat[at].max
minAtValue = stat[at].min
else:
maxAtValue = cl.conditionalDistributions[at].keys()[
len(cl.conditionalDistributions[at].keys()) - 1]
minAtValue = cl.conditionalDistributions[at].keys()[0]
d = maxAtValue - minAtValue
d = getDiff(d / numOfPartitions)
# get curr_num = starting point for continuous att. sampling
curr_num = getStartingPoint(d, minAtValue)
rndFac = getRounding(d)
values = []
for i in range(2 * numOfPartitions):
if curr_num + i * d >= minAtValue and curr_num + i * d <= maxAtValue:
# get thickness
if self.data:
thickness = float(
len(
self.data.filter({
att[at].name:
(curr_num + i * d - d / 2,
curr_num + i * d + d / 2)
}))) / len(self.data)
else:
thickness = 0.0
d_filter = filter(
lambda x: x > curr_num + i * d - d / 2 and x <
curr_num + i * d + d / 2,
cl.conditionalDistributions[at].keys())
if len(d_filter) > 0:
cd = cl.conditionalDistributions[at]
conditional0 = avg([
cd[f][classVal[self.TargetClassIndex]]
for f in d_filter
])
conditional0 = min(1 - aproxZero,
max(aproxZero, conditional0))
conditional1 = 1 - conditional0
try:
# compute error of loess in logistic space
var = avg([
cd[f].variances[self.TargetClassIndex]
for f in d_filter
])
standard_error = math.sqrt(var)
rightError0 = (conditional0 +
standard_error) / max(
conditional1 -
standard_error, aproxZero)
leftError0 = max(conditional0 - standard_error,
aproxZero) / (conditional1 +
standard_error)
se = (math.log(rightError0) -
math.log(leftError0)) / 2
se = math.sqrt(
math.pow(se, 2) + math.pow(priorError, 2))
# add value to set of values
a.addAttValue(
AttValue(str(
round(curr_num + i * d, rndFac)),
math.log(conditional0 /
conditional1 / prior),
lineWidth=thickness,
error=se))
except:
pass
a.continuous = True
# invert values:
# if there are more than 1 value in the attribute, add it to the nomogram
if a and len(a.attValues) > 1:
self.bnomogram.addAttribute(a)
self.alignRadio.setDisabled(False)
self.graph.setScene(self.bnomogram)
self.bnomogram.show()
# Input channel: the logistic regression classifier
def lrClassifier(self, cl):
if self.TargetClassIndex == 0 or self.TargetClassIndex == cl.domain.classVar[
0]:
mult = -1
else:
mult = 1
if self.bnomogram:
self.bnomogram.destroy_and_init(
self, AttValue('Constant', mult * cl.beta[0], error=0))
else:
self.bnomogram = BasicNomogram(
self, AttValue('Constant', mult * cl.beta[0], error=0))
# After applying feature subset selection on discrete attributes
# aproximate unknown error for each attribute is math.sqrt(math.pow(cl.beta_se[0],2)/len(at))
try:
aprox_prior_error = math.sqrt(
math.pow(cl.beta_se[0], 2) / len(cl.domain.attributes))
except:
aprox_prior_error = 0
domain = cl.continuizedDomain or cl.domain
if domain:
for at in domain.attributes:
at.setattr("visited", 0)
for at in domain.attributes:
if at.getValueFrom and at.visited == 0:
name = at.getValueFrom.variable.name
var = at.getValueFrom.variable
if var.ordered:
a = AttrLineOrdered(name, self.bnomogram)
else:
a = AttrLine(name, self.bnomogram)
listOfExcludedValues = []
for val in var.values:
foundValue = False
for same in domain.attributes:
if same.visited == 0 and same.getValueFrom and same.getValueFrom.variable == var and same.getValueFrom.variable.values[
same.getValueFrom.transformer.
value] == val:
same.setattr("visited", 1)
a.addAttValue(
AttValue(val,
mult * cl.beta[same],
error=cl.beta_se[same]))
foundValue = True
if not foundValue:
listOfExcludedValues.append(val)
if len(listOfExcludedValues) == 1:
a.addAttValue(
AttValue(listOfExcludedValues[0],
0,
error=aprox_prior_error))
elif len(listOfExcludedValues) == 2:
a.addAttValue(
AttValue("(" + listOfExcludedValues[0] + "," +
listOfExcludedValues[1] + ")",
0,
error=aprox_prior_error))
elif len(listOfExcludedValues) > 2:
a.addAttValue(
AttValue("Other", 0, error=aprox_prior_error))
# if there are more than 1 value in the attribute, add it to the nomogram
if len(a.attValues) > 1:
self.bnomogram.addAttribute(a)
elif at.visited == 0:
name = at.name
var = at
a = AttrLineCont(name, self.bnomogram)
if self.data:
bas = getCached(self.data, orange.DomainBasicAttrStat,
(self.data, ))
maxAtValue = bas[var].max
minAtValue = bas[var].min
else:
maxAtValue = 1.
minAtValue = -1.
numOfPartitions = 50.
d = getDiff((maxAtValue - minAtValue) / numOfPartitions)
# get curr_num = starting point for continuous att. sampling
curr_num = getStartingPoint(d, minAtValue)
rndFac = getRounding(d)
while curr_num < maxAtValue + d:
if abs(mult * curr_num * cl.beta[at]) < aproxZero:
a.addAttValue(AttValue("0.0", 0))
else:
a.addAttValue(
AttValue(str(curr_num),
mult * curr_num * cl.beta[at]))
curr_num += d
a.continuous = True
at.setattr("visited", 1)
# if there are more than 1 value in the attribute, add it to the nomogram
if len(a.attValues) > 1:
self.bnomogram.addAttribute(a)
self.alignRadio.setDisabled(True)
self.alignType = 0
self.graph.setScene(self.bnomogram)
self.bnomogram.show()
def svmClassifier(self, cl):
import orngLR_Jakulin
import orngLinVis
self.error(0)
if self.TargetClassIndex == 0 or self.TargetClassIndex == cl.domain.classVar[
0]:
mult = -1
else:
mult = 1
try:
visualizer = orngLinVis.Visualizer(self.data,
cl,
buckets=1,
dimensions=1)
beta_from_cl = self.cl.estimator.classifier.classifier.beta[
0] - self.cl.estimator.translator.trans[
0].disp * self.cl.estimator.translator.trans[
0].mult * self.cl.estimator.classifier.classifier.beta[
1]
beta_from_cl = mult * beta_from_cl
except:
self.error(
0, "orngLinVis.Visualizer error" + str(sys.exc_info()[0]) +
":" + str(sys.exc_info()[1]))
# QMessageBox("orngLinVis.Visualizer error", str(sys.exc_info()[0])+":"+str(sys.exc_info()[1]), QMessageBox.Warning,
# QMessageBox.NoButton, QMessageBox.NoButton, QMessageBox.NoButton, self).show()
return
if self.bnomogram:
self.bnomogram.destroy_and_init(
self,
AttValue(
'Constant', -mult * math.log((1.0 / min(
max(visualizer.probfunc(0.0), aproxZero), 0.9999)) -
1), 0))
else:
self.bnomogram = BasicNomogram(
self,
AttValue(
'Constant', -mult * math.log((1.0 / min(
max(visualizer.probfunc(0.0), aproxZero), 0.9999)) -
1), 0))
# get maximum and minimum values in visualizer.m
maxMap = reduce(numpy.maximum, visualizer.m)
minMap = reduce(numpy.minimum, visualizer.m)
coeff = 0 #
at_num = 1
correction = self.cl.coeff * self.cl.estimator.translator.trans[
0].mult * self.cl.estimator.classifier.classifier.beta[1]
for c in visualizer.coeff_names:
if type(c[1]) == str:
for i in range(len(c)):
if i == 0:
if self.data.domain[c[0]].ordered:
a = AttrLineOrdered(c[i], self.bnomogram)
else:
a = AttrLine(c[i], self.bnomogram)
at_num = at_num + 1
else:
if self.data:
thickness = float(
len(
self.data.filter({
self.data.domain[c[0]].name:
str(c[i])
}))) / float(len(self.data))
a.addAttValue(
AttValue(c[i],
correction * mult *
visualizer.coeffs[coeff],
lineWidth=thickness))
coeff = coeff + 1
else:
a = AttrLineCont(c[0], self.bnomogram)
# get min and max from Data and transform coeff accordingly
maxNew = maxMap[coeff]
minNew = maxMap[coeff]
if self.data:
bas = getCached(self.data, orange.DomainBasicAttrStat,
(self.data, ))
maxNew = bas[c[0]].max
minNew = bas[c[0]].min
# transform SVM betas to betas siutable for nomogram
if maxNew == minNew:
beta = ((maxMap[coeff] - minMap[coeff]) /
aproxZero) * visualizer.coeffs[coeff]
else:
beta = ((maxMap[coeff] - minMap[coeff]) /
(maxNew - minNew)) * visualizer.coeffs[coeff]
n = -minNew + minMap[coeff]
numOfPartitions = 50
d = getDiff((maxNew - minNew) / numOfPartitions)
# get curr_num = starting point for continuous att. sampling
curr_num = getStartingPoint(d, minNew)
rndFac = getRounding(d)
while curr_num < maxNew + d:
a.addAttValue(
AttValue(
str(curr_num),
correction *
(mult * (curr_num - minNew) * beta -
minMap[coeff] * visualizer.coeffs[coeff])))
curr_num += d
at_num = at_num + 1
coeff = coeff + 1
a.continuous = True
# if there are more than 1 value in the attribute, add it to the nomogram
if len(a.attValues) > 1:
self.bnomogram.addAttribute(a)
self.cl.domain = orange.Domain(self.data.domain.classVar)
self.graph.setScene(self.bnomogram)
self.bnomogram.show()
# Input channel: the rule classifier (from CN2-EVC only)
def ruleClassifier(self, cl):
def selectSign(oper):
if oper == orange.ValueFilter_continuous.Less:
return "<"
elif oper == orange.ValueFilter_continuous.LessEqual:
return "<="
elif oper == orange.ValueFilter_continuous.Greater:
return ">"
elif oper == orange.ValueFilter_continuous.GreaterEqual:
return ">="
else:
return "="
def getConditions(rule):
conds = rule.filter.conditions
domain = rule.filter.domain
ret = []
if len(conds) == 0:
ret = ret + ["TRUE"]
for i, c in enumerate(conds):
if i > 0:
ret[-1] += " & "
if type(c) == orange.ValueFilter_discrete:
ret += [
domain[c.position].name + "=" +
str(domain[c.position].values[int(c.values[0])])
]
elif type(c) == orange.ValueFilter_continuous:
ret += [
domain[c.position].name + selectSign(c.oper) +
"%.3f" % c.ref
]
return ret
self.error(1)
if not len(self.data.domain.classVar.values) == 2:
self.error(1, "Rules require binary classes")
classVal = cl.domain.classVar
att = cl.domain.attributes
if self.TargetClassIndex == 0 or self.TargetClassIndex == cl.domain.classVar[
0]:
mult = 1.
else:
mult = -1.
# calculate prior probability (from self.TargetClassIndex)
if self.bnomogram:
self.bnomogram.destroy_and_init(self, AttValue("Constant", 0.0))
else:
self.bnomogram = BasicNomogram(self, AttValue("Constant", 0.0))
self.cl.setattr("rulesOrdering", [])
for r_i, r in enumerate(cl.rules):
a = AttrLine(getConditions(r), self.bnomogram)
self.cl.rulesOrdering.append(getConditions(r))
if r.classifier.defaultVal == 0:
sign = mult
else:
sign = -mult
a.addAttValue(
AttValue("yes",
sign * cl.ruleBetas[r_i],
lineWidth=0,
error=0.0))
a.addAttValue(AttValue("no", 0.0, lineWidth=0, error=0.0))
self.bnomogram.addAttribute(a)
self.graph.setScene(self.bnomogram)
self.bnomogram.show()
def initClassValues(self, classValue):
self.targetCombo.clear()
self.targetCombo.addItems([str(v) for v in classValue])
def classifier(self, cl):
self.closeContext()
self.error(2)
oldcl = self.cl
self.cl = None
if cl:
for acceptable in (orange.BayesClassifier, orange.LogRegClassifier,
orange.RuleClassifier_logit):
if isinstance(cl, acceptable):
self.cl = cl
break
else:
self.error(
2,
"Nomograms can be drawn for only for Bayesian classifier and logistic regression"
)
if not oldcl or not self.cl or not oldcl.domain == self.cl.domain:
if self.cl:
self.initClassValues(self.cl.domain.classVar)
self.TargetClassIndex = 0
self.data = getattr(self.cl, "data", None)
if self.data and self.data.domain and not self.data.domain.classVar:
self.error(2, "Classless domain")
# Here it said "return", but let us report an error and clean up the widget
self.cl = self.data = None
self.openContext("", self.data)
if not self.data:
self.histogramCheck.setChecked(False)
self.histogramCheck.setDisabled(True)
self.histogramLabel.setDisabled(True)
self.CICheck.setChecked(False)
self.CICheck.setDisabled(True)
self.CILabel.setDisabled(True)
else:
self.histogramCheck.setEnabled(True)
self.histogramCheck.makeConsistent()
self.CICheck.setEnabled(True)
self.CICheck.makeConsistent()
self.updateNomogram()
def setTarget(self):
self.updateNomogram()
def updateNomogram(self):
## import orngSVM
def setNone():
for view in [self.footer, self.header, self.graph]:
scene = view.scene()
if scene:
for item in scene.items():
scene.removeItem(item)
if self.data and self.cl: # and not type(self.cl) == orngLR_Jakulin.MarginMetaClassifier:
#check domains
for at in self.cl.domain:
if at.getValueFrom and hasattr(at.getValueFrom, "variable"):
if (not at.getValueFrom.variable in self.data.domain) and (
not at in self.data.domain):
return
else:
if not at in self.data.domain:
return
if isinstance(self.cl, orange.BayesClassifier):
# if len(self.cl.domain.classVar.values)>2:
# QMessageBox("OWNomogram:", " Please use only Bayes classifiers that are induced on data with dichotomous class!", QMessageBox.Warning,
# QMessageBox.NoButton, QMessageBox.NoButton, QMessageBox.NoButton, self).show()
# else:
self.nbClassifier(self.cl)
## elif isinstance(self.cl, orngLR_Jakulin.MarginMetaClassifier) and self.data:
## self.svmClassifier(self.cl)
elif isinstance(self.cl, orange.RuleClassifier_logit):
self.ruleClassifier(self.cl)
elif isinstance(self.cl, orange.LogRegClassifier):
# get if there are any continuous attributes in data -> then we need data to compute margins
cont = False
if self.cl.continuizedDomain:
for at in self.cl.continuizedDomain.attributes:
if not at.getValueFrom:
cont = True
if self.data or not cont:
self.lrClassifier(self.cl)
else:
setNone()
else:
setNone()
if self.sort_type > 0:
self.sortNomogram()
def sortNomogram(self):
def sign(x):
if x < 0:
return -1
return 1
def compare_to_ordering_in_rules(x, y):
return self.cl.rulesOrdering.index(
x.name) - self.cl.rulesOrdering.index(y.name)
def compare_to_ordering_in_data(x, y):
return self.data.domain.attributes.index(
self.data.domain[x.name]) - self.data.domain.attributes.index(
self.data.domain[y.name])
def compare_to_ordering_in_domain(x, y):
return self.cl.domain.attributes.index(
self.cl.domain[x.name]) - self.cl.domain.attributes.index(
self.cl.domain[y.name])
def compate_beta_difference(x, y):
return -sign(x.maxValue - x.minValue - y.maxValue + y.minValue)
def compare_beta_positive(x, y):
return -sign(x.maxValue - y.maxValue)
def compare_beta_negative(x, y):
return sign(x.minValue - y.minValue)
if not self.bnomogram:
return
if self.sort_type == 0 and hasattr(self.cl, "rulesOrdering"):
self.bnomogram.attributes.sort(compare_to_ordering_in_rules)
elif self.sort_type == 0 and self.data:
self.bnomogram.attributes.sort(compare_to_ordering_in_data)
elif self.sort_type == 0 and self.cl and self.cl.domain:
self.bnomogram.attributes.sort(compare_to_ordering_in_domain)
if self.sort_type == 1:
self.bnomogram.attributes.sort(compate_beta_difference)
elif self.sort_type == 2:
self.bnomogram.attributes.sort(compare_beta_positive)
elif self.sort_type == 3:
self.bnomogram.attributes.sort(compare_beta_negative)
# update nomogram
self.showNomogram()
def setProbability(self):
if self.probability and self.bnomogram:
self.bnomogram.showAllMarkers()
elif self.bnomogram:
self.bnomogram.hideAllMarkers()
def setBaseLine(self):
if self.bnomogram:
self.bnomogram.showBaseLine(True)
def menuItemPrinter(self):
import copy
canvases = header, graph, footer = self.header.scene(
), self.graph.scene(), self.footer.scene()
# all scenes together
scene_confed = QGraphicsScene(0, 0, max(c.width() for c in canvases),
sum(c.height() for c in canvases))
# add items from header
header_its = header.items()
for it in header_its:
scene_confed.addItem(it)
# add items from graph
graph_its = graph.items()
for it in graph_its:
scene_confed.addItem(it)
it.moveBy(0., header.height())
# add from footer
footer_its = footer.items()
for it in footer_its:
scene_confed.addItem(it)
it.moveBy(0., header.height() + graph.height())
try:
import OWDlgs
except:
print "Missing file 'OWDlgs.py'. This file should be in OrangeWidgets folder. Unable to print/save image."
sizeDlg = OWDlgs.OWChooseImageSizeDlg(scene_confed, parent=self)
sizeDlg.exec_()
# set all items back to original canvases
for it in header_its:
header.addItem(it)
for it in graph_its:
graph.addItem(it)
it.moveBy(0., -header.height())
for it in footer_its:
footer.addItem(it)
it.moveBy(0, -header.height() - graph.height())
self.showNomogram()
# Callbacks
def showNomogram(self):
if self.bnomogram and self.cl:
#self.bnomogram.hide()
self.bnomogram.show()
self.bnomogram.update()
class OWNomogram(OWWidget):
settingsList = ["alignType", "verticalSpacing", "contType", "verticalSpacingContinuous", "yAxis", "probability", "confidence_check", "confidence_percent", "histogram", "histogram_size", "sort_type"]
contextHandlers = {"": DomainContextHandler("", ["TargetClassIndex"], matchValues=1)}
def __init__(self,parent=None, signalManager = None):
OWWidget.__init__(self, parent, signalManager, "Nomogram", 1)
#self.setWFlags(Qt.WResizeNoErase | Qt.WRepaintNoErase) #this works like magic.. no flicker during repaint!
self.parent = parent
# self.setWFlags(self.getWFlags()+Qt.WStyle_Maximize)
self.callbackDeposit = [] # deposit for OWGUI callback functions
self.alignType = 0
self.contType = 0
self.yAxis = 0
self.probability = 0
self.verticalSpacing = 60
self.verticalSpacingContinuous = 100
self.diff_between_ordinal = 30
self.fontSize = 9
self.lineWidth = 1
self.histogram = 0
self.histogram_size = 10
self.data = None
self.cl = None
self.confidence_check = 0
self.confidence_percent = 95
self.sort_type = 0
self.loadSettings()
self.pointsName = ["Total", "Total"]
self.totalPointsName = ["Probability", "Probability"]
self.bnomogram = None
self.inputs=[("Classifier", orange.Classifier, self.classifier)]
self.TargetClassIndex = 0
self.targetCombo = OWGUI.comboBox(self.controlArea, self, "TargetClassIndex", " Target Class ", addSpace=True, tooltip='Select target (prediction) class in the model.', callback = self.setTarget)
self.alignRadio = OWGUI.radioButtonsInBox(self.controlArea, self, 'alignType', ['Align left', 'Align by zero influence'], box='Attribute placement',
tooltips=['Attributes in nomogram are left aligned', 'Attributes are not aligned, top scale represents true (normalized) regression coefficient value'],
addSpace=True,
callback=self.showNomogram)
self.verticalSpacingLabel = OWGUI.spin(self.alignRadio, self, 'verticalSpacing', 15, 200, label = 'Vertical spacing:', orientation = 0, tooltip='Define space (pixels) between adjacent attributes.', callback = self.showNomogram)
self.ContRadio = OWGUI.radioButtonsInBox(self.controlArea, self, 'contType', ['1D projection', '2D curve'], 'Continuous attributes',
tooltips=['Continuous attribute are presented on a single scale', 'Two dimensional space is used to present continuous attributes in nomogram.'],
addSpace=True,
callback=[lambda:self.verticalSpacingContLabel.setDisabled(not self.contType), self.showNomogram])
self.verticalSpacingContLabel = OWGUI.spin(OWGUI.indentedBox(self.ContRadio, sep=OWGUI.checkButtonOffsetHint(self.ContRadio.buttons[-1])), self, 'verticalSpacingContinuous', 15, 200, label = "Height", orientation=0, tooltip='Define space (pixels) between adjacent 2d presentation of attributes.', callback = self.showNomogram)
self.verticalSpacingContLabel.setDisabled(not self.contType)
self.yAxisRadio = OWGUI.radioButtonsInBox(self.controlArea, self, 'yAxis', ['Point scale', 'Log odds ratios'], 'Scale',
tooltips=['values are normalized on a 0-100 point scale','values on top axis show log-linear contribution of attribute to full model'],
addSpace=True,
callback=self.showNomogram)
layoutBox = OWGUI.widgetBox(self.controlArea, "Display", orientation=1, addSpace=True)
self.probabilityCheck = OWGUI.checkBox(layoutBox, self, 'probability', 'Show prediction', tooltip='', callback = self.setProbability)
self.CICheck, self.CILabel = OWGUI.checkWithSpin(layoutBox, self, 'Confidence intervals (%):', min=1, max=99, step = 1, checked='confidence_check', value='confidence_percent', checkCallback=self.showNomogram, spinCallback = self.showNomogram)
self.histogramCheck, self.histogramLabel = OWGUI.checkWithSpin(layoutBox, self, 'Show histogram, size', min=1, max=30, checked='histogram', value='histogram_size', step = 1, tooltip='-(TODO)-', checkCallback=self.showNomogram, spinCallback = self.showNomogram)
OWGUI.separator(layoutBox)
self.sortOptions = ["No sorting", "Absolute importance", "Positive influence", "Negative influence"]
self.sortBox = OWGUI.comboBox(layoutBox, self, "sort_type", label="Sort by ", items=self.sortOptions, callback = self.sortNomogram, orientation="horizontal")
OWGUI.rubber(self.controlArea)
self.connect(self.graphButton, SIGNAL("clicked()"), self.menuItemPrinter)
#add a graph widget
self.header = OWNomogramHeader(None, self.mainArea)
self.header.setFixedHeight(60)
self.header.setVerticalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.header.setHorizontalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.graph = OWNomogramGraph(self.bnomogram, self.mainArea)
self.graph.setMinimumWidth(200)
self.graph.setHorizontalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.footer = OWNomogramHeader(None, self.mainArea)
self.footer.setFixedHeight(60*2+10)
self.footer.setVerticalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.footer.setHorizontalScrollBarPolicy(Qt.ScrollBarAlwaysOff)
self.mainArea.layout().addWidget(self.header)
self.mainArea.layout().addWidget(self.graph)
self.mainArea.layout().addWidget(self.footer)
self.resize(700,500)
#self.repaint()
#self.update()
# mouse pressed flag
self.mousepr = False
def sendReport(self):
if self.cl:
tclass = self.cl.domain.classVar.values[self.TargetClassIndex]
else:
tclass = "N/A"
self.reportSettings("Information",
[("Target class", tclass),
self.confidence_check and ("Confidence intervals", "%i%%" % self.confidence_percent),
("Sorting", self.sortOptions[self.sort_type] if self.sort_type else "None")])
if self.cl:
canvases = header, graph, footer = self.header.scene(), self.graph.scene(), self.footer.scene()
painter = QPainter()
buffer = QPixmap(max(c.width() for c in canvases), sum(c.height() for c in canvases))
painter.begin(buffer)
painter.fillRect(buffer.rect(), QBrush(QColor(255, 255, 255)))
header.render(painter, QRectF(0, 0, header.width(), header.height()), QRectF(0, 0, header.width(), header.height()))
graph.render(painter, QRectF(0, header.height(), graph.width(), graph.height()), QRectF(0, 0, graph.width(), graph.height()))
footer.render(painter, QRectF(0, header.height()+graph.height(), footer.width(), footer.height()), QRectF(0, 0, footer.width(), footer.height()))
painter.end()
self.reportImage(lambda filename: buffer.save(filename, os.path.splitext(filename)[1][1:]))
# Input channel: the Bayesian classifier
def nbClassifier(self, cl):
# this subroutine computes standard error of estimated beta. Note that it is used only for discrete data,
# continuous data have a different computation.
def errOld(e, priorError, key, data):
inf = 0.0
sume = e[0]+e[1]
for d in data:
if d[at]==key:
inf += (e[0]*e[1]/sume/sume)
inf = max(inf, aproxZero)
var = max(1/inf - priorError*priorError, 0)
return (math.sqrt(var))
def err(condDist, att, value, targetClass, priorError, data):
sumE = sum(condDist)
valueE = condDist[targetClass]
distAtt = orange.Distribution(att, data)
inf = distAtt[value]*(valueE/sumE)*(1-valueE/sumE)
inf = max(inf, aproxZero)
var = max(1/inf - priorError*priorError, 0)
return (math.sqrt(var))
classVal = cl.domain.classVar
att = cl.domain.attributes
# calculate prior probability
dist1 = max(aproxZero, 1-cl.distribution[classVal[self.TargetClassIndex]])
dist0 = max(aproxZero, cl.distribution[classVal[self.TargetClassIndex]])
prior = dist0/dist1
if self.data:
sumd = dist1+dist0
priorError = math.sqrt(1/((dist1*dist0/sumd/sumd)*len(self.data)))
else:
priorError = 0
if self.bnomogram:
self.bnomogram.destroy_and_init(self, AttValue("Constant", math.log(prior), error = priorError))
else:
self.bnomogram = BasicNomogram(self, AttValue("Constant", math.log(prior), error = priorError))
if self.data:
stat = getCached(self.data, orange.DomainBasicAttrStat, (self.data,))
for at in range(len(att)):
a = None
if att[at].varType == orange.VarTypes.Discrete:
if att[at].ordered:
a = AttrLineOrdered(att[at].name, self.bnomogram)
else:
a = AttrLine(att[at].name, self.bnomogram)
for cd in cl.conditionalDistributions[at].keys():
# calculuate thickness
conditional0 = max(cl.conditionalDistributions[at][cd][classVal[self.TargetClassIndex]], aproxZero)
conditional1 = max(1-cl.conditionalDistributions[at][cd][classVal[self.TargetClassIndex]], aproxZero)
beta = math.log(conditional0/conditional1/prior)
if self.data:
#thickness = int(round(4.*float(len(self.data.filter({att[at].name:str(cd)})))/float(len(self.data))))
thickness = float(len(self.data.filter({att[at].name:str(cd)})))/float(len(self.data))
se = err(cl.conditionalDistributions[at][cd], att[at], cd, classVal[self.TargetClassIndex], priorError, self.data) # standar error of beta
else:
thickness = 0
se = 0
a.addAttValue(AttValue(str(cd), beta, lineWidth=thickness, error = se))
else:
a = AttrLineCont(att[at].name, self.bnomogram)
numOfPartitions = 50
if self.data:
maxAtValue = stat[at].max
minAtValue = stat[at].min
else:
maxAtValue = cl.conditionalDistributions[at].keys()[len(cl.conditionalDistributions[at].keys())-1]
minAtValue = cl.conditionalDistributions[at].keys()[0]
d = maxAtValue-minAtValue
d = getDiff(d/numOfPartitions)
# get curr_num = starting point for continuous att. sampling
curr_num = getStartingPoint(d, minAtValue)
rndFac = getRounding(d)
values = []
for i in range(2*numOfPartitions):
if curr_num+i*d>=minAtValue and curr_num+i*d<=maxAtValue:
# get thickness
if self.data:
thickness = float(len(self.data.filter({att[at].name:(curr_num+i*d-d/2, curr_num+i*d+d/2)})))/len(self.data)
else:
thickness = 0.0
d_filter = filter(lambda x: x>curr_num+i*d-d/2 and x<curr_num+i*d+d/2, cl.conditionalDistributions[at].keys())
if len(d_filter)>0:
cd = cl.conditionalDistributions[at]
conditional0 = avg([cd[f][classVal[self.TargetClassIndex]] for f in d_filter])
conditional0 = min(1-aproxZero,max(aproxZero,conditional0))
conditional1 = 1-conditional0
try:
# compute error of loess in logistic space
var = avg([cd[f].variances[self.TargetClassIndex] for f in d_filter])
standard_error= math.sqrt(var)
rightError0 = (conditional0+standard_error)/max(conditional1-standard_error, aproxZero)
leftError0 = max(conditional0-standard_error, aproxZero)/(conditional1+standard_error)
se = (math.log(rightError0) - math.log(leftError0))/2
se = math.sqrt(math.pow(se,2)+math.pow(priorError,2))
# add value to set of values
a.addAttValue(AttValue(str(round(curr_num+i*d,rndFac)),
math.log(conditional0/conditional1/prior),
lineWidth=thickness,
error = se))
except:
pass
a.continuous = True
# invert values:
# if there are more than 1 value in the attribute, add it to the nomogram
if a and len(a.attValues)>1:
self.bnomogram.addAttribute(a)
self.alignRadio.setDisabled(False)
self.graph.setScene(self.bnomogram)
self.bnomogram.show()
# Input channel: the logistic regression classifier
def lrClassifier(self, cl):
if self.TargetClassIndex == 0 or self.TargetClassIndex == cl.domain.classVar[0]:
mult = -1
else:
mult = 1
if self.bnomogram:
self.bnomogram.destroy_and_init(self, AttValue('Constant', mult*cl.beta[0], error = 0))
else:
self.bnomogram = BasicNomogram(self, AttValue('Constant', mult*cl.beta[0], error = 0))
# After applying feature subset selection on discrete attributes
# aproximate unknown error for each attribute is math.sqrt(math.pow(cl.beta_se[0],2)/len(at))
try:
aprox_prior_error = math.sqrt(math.pow(cl.beta_se[0],2)/len(cl.domain.attributes))
except:
aprox_prior_error = 0
domain = cl.continuizedDomain or cl.domain
if domain:
for at in domain.attributes:
at.setattr("visited",0)
for at in domain.attributes:
if at.getValueFrom and at.visited==0:
name = at.getValueFrom.variable.name
var = at.getValueFrom.variable
if var.ordered:
a = AttrLineOrdered(name, self.bnomogram)
else:
a = AttrLine(name, self.bnomogram)
listOfExcludedValues = []
for val in var.values:
foundValue = False
for same in domain.attributes:
if same.visited==0 and same.getValueFrom and same.getValueFrom.variable == var and same.getValueFrom.variable.values[same.getValueFrom.transformer.value]==val:
same.setattr("visited",1)
a.addAttValue(AttValue(val, mult*cl.beta[same], error = cl.beta_se[same]))
foundValue = True
if not foundValue:
listOfExcludedValues.append(val)
if len(listOfExcludedValues) == 1:
a.addAttValue(AttValue(listOfExcludedValues[0], 0, error = aprox_prior_error))
elif len(listOfExcludedValues) == 2:
a.addAttValue(AttValue("("+listOfExcludedValues[0]+","+listOfExcludedValues[1]+")", 0, error = aprox_prior_error))
elif len(listOfExcludedValues) > 2:
a.addAttValue(AttValue("Other", 0, error = aprox_prior_error))
# if there are more than 1 value in the attribute, add it to the nomogram
if len(a.attValues)>1:
self.bnomogram.addAttribute(a)
elif at.visited==0:
name = at.name
var = at
a = AttrLineCont(name, self.bnomogram)
if self.data:
bas = getCached(self.data, orange.DomainBasicAttrStat, (self.data,))
maxAtValue = bas[var].max
minAtValue = bas[var].min
else:
maxAtValue = 1.
minAtValue = -1.
numOfPartitions = 50.
d = getDiff((maxAtValue-minAtValue)/numOfPartitions)
# get curr_num = starting point for continuous att. sampling
curr_num = getStartingPoint(d, minAtValue)
rndFac = getRounding(d)
while curr_num<maxAtValue+d:
if abs(mult*curr_num*cl.beta[at])<aproxZero:
a.addAttValue(AttValue("0.0", 0))
else:
a.addAttValue(AttValue(str(curr_num), mult*curr_num*cl.beta[at]))
curr_num += d
a.continuous = True
at.setattr("visited", 1)
# if there are more than 1 value in the attribute, add it to the nomogram
if len(a.attValues)>1:
self.bnomogram.addAttribute(a)
self.alignRadio.setDisabled(True)
self.alignType = 0
self.graph.setScene(self.bnomogram)
self.bnomogram.show()
def lr2Classifier(self, cl):
if len(cl.weights) == 1:
weights = list(cl.weights[0])
if self.TargetClassIndex > 0:
weights = [ -w for w in weights ]
else:
weights = list(cl.weights[self.TargetClassIndex])
domain = cl.domain
for at in domain.attributes:
at.setattr("visited",0)
order = []
multiple = defaultdict(list)
for iat,at in enumerate(domain.attributes):
if at.getValueFrom and hasattr(at.getValueFrom, "variable") \
and isinstance(at.getValueFrom.variable, Orange.feature.Discrete):
var = at.getValueFrom.variable
if var not in multiple:
order.append((var,-1))
multiple[var].append((at, iat))
else:
order.append((at, iat)) #raw variable
candidateats = []
for at,iat in order:
if at not in multiple:
span = 1.
average = 0.
basevar = at
#check if normalized
if isinstance(at.getValueFrom, Orange.classification.ClassifierFromVar) \
and isinstance(at.getValueFrom.variable, Orange.feature.Continuous) \
and isinstance(at.getValueFrom.transformer, Orange.core.NormalizeContinuous):
span = at.getValueFrom.transformer.span
average = at.getValueFrom.transformer.average
basevar = at.getValueFrom.variable
name = basevar.name
if self.data:
bas = getCached(self.data, orange.DomainBasicAttrStat, (self.data,))
maxAtValue = bas[basevar].max
minAtValue = bas[basevar].min
avgAtValue = bas[basevar].avg
else:
maxAtValue = 1.
minAtValue = -1.
avgAtValue = 0.
# get curr_num = starting point for continuous att. sampling
w = weights[iat]
betas = [ ((minAtValue-average)/span)*w, ((maxAtValue-average)/span)*w ]
rng = max(betas) - min(betas)
freqbase = ((avgAtValue-average)/span)*w
candidateats.append((at, name, "cont", minAtValue, maxAtValue, w, average, span, freqbase, rng))
else:
found_values = set()
values = []
for ati, iati in multiple[at]:
val = ati.getValueFrom.variable.values[ati.getValueFrom.transformer.value]
values.append((val, weights[iati]))
found_values.add(val)
excluded_values = list(set(at.values) - found_values)
excluded_name = excluded_values[0] if len(excluded_values) == 1 else "other"
values.append((excluded_name, 0))
betas = [ a[1] for a in values ]
rng = max(betas) - min(betas)
#the most frequent item has weight 0.
candidateats.append((at, at.name, "disc", values, 0., rng))
self.warning(1011)
biggest_influence = set([a[0] for a in sorted(candidateats, key=lambda x: -x[-1]) ][:SHOW_BIGGEST])
if len(biggest_influence) < len(candidateats):
self.warning(1011, "Showing only %d features with highest influence (out of %d)." % (len(biggest_influence), len(candidateats) ))
hiddenbeta = 0.
for aaa in candidateats:
if aaa[0] not in biggest_influence:
hiddenbeta += aaa[-2]
if self.bnomogram:
self.bnomogram.destroy_and_init(self, AttValue('Constant', weights[-1]+hiddenbeta))
else:
self.bnomogram = BasicNomogram(self, AttValue('Constant', weights[-1]+hiddenbeta))
for aaa in candidateats:
if aaa[0] in biggest_influence:
if aaa[2] == "cont":
_, name, _, minAtValue, maxAtValue, w, average, span, _, _ = aaa
numOfPartitions = 50.
d = getDiff((maxAtValue-minAtValue)/numOfPartitions)
a = AttrLineCont(name, self.bnomogram)
curr_num = getStartingPoint(d, minAtValue)
while curr_num<maxAtValue+d:
if abs(curr_num*w)<aproxZero:
a.addAttValue(AttValue("0.0", 0))
else:
a.addAttValue(AttValue(str(curr_num), ((curr_num-average)/span)*w))
curr_num += d
a.continuous = True
elif aaa[2] == "disc":
_, name, _, values, _, _ = aaa
a = AttrLine(name, self.bnomogram)
for n, val in values:
a.addAttValue(AttValue(n, val))
if len(a.attValues)>1:
self.bnomogram.addAttribute(a)
self.alignRadio.setDisabled(True)
self.alignType = 0
self.graph.setScene(self.bnomogram)
self.bnomogram.show()
def svmClassifier(self, cl):
import orngLR_Jakulin
import orngLinVis
self.error(0)
if self.TargetClassIndex == 0 or self.TargetClassIndex == cl.domain.classVar[0]:
mult = -1
else:
mult = 1
try:
visualizer = orngLinVis.Visualizer(self.data, cl, buckets=1, dimensions=1)
beta_from_cl = self.cl.estimator.classifier.classifier.beta[0] - self.cl.estimator.translator.trans[0].disp*self.cl.estimator.translator.trans[0].mult*self.cl.estimator.classifier.classifier.beta[1]
beta_from_cl = mult*beta_from_cl
except:
self.error(0, "orngLinVis.Visualizer error"+ str(sys.exc_info()[0])+":"+str(sys.exc_info()[1]))
# QMessageBox("orngLinVis.Visualizer error", str(sys.exc_info()[0])+":"+str(sys.exc_info()[1]), QMessageBox.Warning,
# QMessageBox.NoButton, QMessageBox.NoButton, QMessageBox.NoButton, self).show()
return
if self.bnomogram:
self.bnomogram.destroy_and_init(self, AttValue('Constant', -mult*math.log((1.0/min(max(visualizer.probfunc(0.0),aproxZero),0.9999))-1), 0))
else:
self.bnomogram = BasicNomogram(self, AttValue('Constant', -mult*math.log((1.0/min(max(visualizer.probfunc(0.0),aproxZero),0.9999))-1), 0))
# get maximum and minimum values in visualizer.m
maxMap = reduce(numpy.maximum, visualizer.m)
minMap = reduce(numpy.minimum, visualizer.m)
coeff = 0 #
at_num = 1
correction = self.cl.coeff*self.cl.estimator.translator.trans[0].mult*self.cl.estimator.classifier.classifier.beta[1]
for c in visualizer.coeff_names:
if type(c[1])==str:
for i in range(len(c)):
if i == 0:
if self.data.domain[c[0]].ordered:
a = AttrLineOrdered(c[i], self.bnomogram)
else:
a = AttrLine(c[i], self.bnomogram)
at_num = at_num + 1
else:
if self.data:
thickness = float(len(self.data.filter({self.data.domain[c[0]].name:str(c[i])})))/float(len(self.data))
a.addAttValue(AttValue(c[i], correction*mult*visualizer.coeffs[coeff], lineWidth=thickness))
coeff = coeff + 1
else:
a = AttrLineCont(c[0], self.bnomogram)
# get min and max from Data and transform coeff accordingly
maxNew=maxMap[coeff]
minNew=maxMap[coeff]
if self.data:
bas = getCached(self.data, orange.DomainBasicAttrStat, (self.data,))
maxNew = bas[c[0]].max
minNew = bas[c[0]].min
# transform SVM betas to betas siutable for nomogram
if maxNew == minNew:
beta = ((maxMap[coeff]-minMap[coeff])/aproxZero)*visualizer.coeffs[coeff]
else:
beta = ((maxMap[coeff]-minMap[coeff])/(maxNew-minNew))*visualizer.coeffs[coeff]
n = -minNew+minMap[coeff]
numOfPartitions = 50
d = getDiff((maxNew-minNew)/numOfPartitions)
# get curr_num = starting point for continuous att. sampling
curr_num = getStartingPoint(d, minNew)
rndFac = getRounding(d)
while curr_num<maxNew+d:
a.addAttValue(AttValue(str(curr_num), correction*(mult*(curr_num-minNew)*beta-minMap[coeff]*visualizer.coeffs[coeff])))
curr_num += d
at_num = at_num + 1
coeff = coeff + 1
a.continuous = True
# if there are more than 1 value in the attribute, add it to the nomogram
if len(a.attValues)>1:
self.bnomogram.addAttribute(a)
self.cl.domain = orange.Domain(self.data.domain.classVar)
self.graph.setScene(self.bnomogram)
self.bnomogram.show()
# Input channel: the rule classifier (from CN2-EVC only)
def ruleClassifier(self, cl):
def selectSign(oper):
if oper == orange.ValueFilter_continuous.Less:
return "<"
elif oper == orange.ValueFilter_continuous.LessEqual:
return "<="
elif oper == orange.ValueFilter_continuous.Greater:
return ">"
elif oper == orange.ValueFilter_continuous.GreaterEqual:
return ">="
else: return "="
def getConditions(rule):
conds = rule.filter.conditions
domain = rule.filter.domain
ret = []
if len(conds)==0:
ret = ret + ["TRUE"]
for i,c in enumerate(conds):
if i > 0:
ret[-1] += " & "
if type(c) == orange.ValueFilter_discrete:
ret += [domain[c.position].name + "=" + str(domain[c.position].values[int(c.values[0])])]
elif type(c) == orange.ValueFilter_continuous:
ret += [domain[c.position].name + selectSign(c.oper) + "%.3f"%c.ref]
return ret
self.error(1)
if not len(self.data.domain.classVar.values) == 2:
self.error(1, "Rules require binary classes")
classVal = cl.domain.classVar
att = cl.domain.attributes
if self.TargetClassIndex == 0 or self.TargetClassIndex == cl.domain.classVar[0]:
mult = 1.
else:
mult = -1.
# calculate prior probability (from self.TargetClassIndex)
if self.bnomogram:
self.bnomogram.destroy_and_init(self, AttValue("Constant", 0.0))
else:
self.bnomogram = BasicNomogram(self, AttValue("Constant", 0.0))
self.cl.setattr("rulesOrdering", [])
for r_i,r in enumerate(cl.rules):
a = AttrLine(getConditions(r), self.bnomogram)
self.cl.rulesOrdering.append(getConditions(r))
if r.classifier.defaultVal == 0:
sign = mult
else: sign = -mult
a.addAttValue(AttValue("yes", sign*cl.ruleBetas[r_i], lineWidth=0, error = 0.0))
a.addAttValue(AttValue("no", 0.0, lineWidth=0, error = 0.0))
self.bnomogram.addAttribute(a)
self.graph.setScene(self.bnomogram)
self.bnomogram.show()
def initClassValues(self, classValue):
self.targetCombo.clear()
self.targetCombo.addItems([str(v) for v in classValue])
def classifier(self, cl):
self.closeContext()
self.error(2)
oldcl = self.cl
self.cl = None
if cl:
for acceptable in (orange.BayesClassifier, orange.LogRegClassifier, orange.RuleClassifier_logit, Orange.classification.svm.LinearClassifier):
if isinstance(cl, acceptable):
self.cl = cl
break
else:
self.error(2, "Nomograms can be drawn for only for Bayesian classifier and logistic regression")
if not oldcl or not self.cl or not oldcl.domain == self.cl.domain:
if self.cl:
self.initClassValues(self.cl.domain.classVar)
self.TargetClassIndex = 0
self.data = getattr(self.cl, "data", None)
if self.data and self.data.domain and not self.data.domain.classVar:
self.error(2, "Classless domain")
# Here it said "return", but let us report an error and clean up the widget
self.cl = self.data = None
self.openContext("", self.data)
if not self.data:
self.histogramCheck.setChecked(False)
self.histogramCheck.setDisabled(True)
self.histogramLabel.setDisabled(True)
self.CICheck.setChecked(False)
self.CICheck.setDisabled(True)
self.CILabel.setDisabled(True)
else:
self.histogramCheck.setEnabled(True)
self.histogramCheck.makeConsistent()
self.CICheck.setEnabled(True)
self.CICheck.makeConsistent()
self.updateNomogram()
def setTarget(self):
self.updateNomogram()
def updateNomogram(self):
## import orngSVM
def setNone():
for view in [self.footer, self.header, self.graph]:
scene = view.scene()
if scene:
for item in scene.items():
scene.removeItem(item)
if self.data and self.cl: # and not type(self.cl) == orngLR_Jakulin.MarginMetaClassifier:
#check domains
for at in self.cl.domain:
if at.getValueFrom and hasattr(at.getValueFrom, "variable"):
if (not at.getValueFrom.variable in self.data.domain) and (not at in self.data.domain):
return
else:
if not at in self.data.domain:
return
if isinstance(self.cl, orange.BayesClassifier):
# if len(self.cl.domain.classVar.values)>2:
# QMessageBox("OWNomogram:", " Please use only Bayes classifiers that are induced on data with dichotomous class!", QMessageBox.Warning,
# QMessageBox.NoButton, QMessageBox.NoButton, QMessageBox.NoButton, self).show()
# else:
self.nbClassifier(self.cl)
## elif isinstance(self.cl, orngLR_Jakulin.MarginMetaClassifier) and self.data:
## self.svmClassifier(self.cl)
elif isinstance(self.cl, orange.RuleClassifier_logit):
self.ruleClassifier(self.cl)
elif isinstance(self.cl, orange.LogRegClassifier):
# get if there are any continuous attributes in data -> then we need data to compute margins
cont = False
if self.cl.continuizedDomain:
for at in self.cl.continuizedDomain.attributes:
if not at.getValueFrom:
cont = True
if self.data or not cont:
self.lrClassifier(self.cl)
else:
setNone()
elif isinstance(self.cl, Orange.classification.svm.LinearClassifier):
self.lr2Classifier(self.cl)
else:
setNone()
if self.sort_type>0:
self.sortNomogram()
def sortNomogram(self):
def sign(x):
if x<0:
return -1;
return 1;
def compare_to_ordering_in_rules(x,y):
return self.cl.rulesOrdering.index(x.name) - self.cl.rulesOrdering.index(y.name)
def compare_to_ordering_in_data(x,y):
return self.data.domain.attributes.index(self.data.domain[x.name]) - self.data.domain.attributes.index(self.data.domain[y.name])
def compare_to_ordering_in_domain(x,y):
return self.cl.domain.attributes.index(self.cl.domain[x.name]) - self.cl.domain.attributes.index(self.cl.domain[y.name])
def compate_beta_difference(x,y):
return -sign(x.maxValue-x.minValue-y.maxValue+y.minValue)
def compare_beta_positive(x, y):
return -sign(x.maxValue-y.maxValue)
def compare_beta_negative(x, y):
return sign(x.minValue-y.minValue)
if not self.bnomogram:
return
if self.sort_type == 0 and hasattr(self.cl, "rulesOrdering"):
self.bnomogram.attributes.sort(compare_to_ordering_in_rules)
elif self.sort_type == 0 and self.data:
self.bnomogram.attributes.sort(compare_to_ordering_in_data)
elif self.sort_type == 0 and self.cl and self.cl.domain:
self.bnomogram.attributes.sort(compare_to_ordering_in_domain)
if self.sort_type == 1:
self.bnomogram.attributes.sort(compate_beta_difference)
elif self.sort_type == 2:
self.bnomogram.attributes.sort(compare_beta_positive)
elif self.sort_type == 3:
self.bnomogram.attributes.sort(compare_beta_negative)
# update nomogram
self.showNomogram()
def setProbability(self):
if self.probability and self.bnomogram:
self.bnomogram.showAllMarkers()
elif self.bnomogram:
self.bnomogram.hideAllMarkers()
def setBaseLine(self):
if self.bnomogram:
self.bnomogram.showBaseLine(True)
def menuItemPrinter(self):
import copy
canvases = header, graph, footer = self.header.scene(), self.graph.scene(), self.footer.scene()
# all scenes together
scene_confed = QGraphicsScene(0, 0, max(c.width() for c in canvases), sum(c.height() for c in canvases))
# add items from header
header_its = header.items()
for it in header_its:
scene_confed.addItem(it)
# add items from graph
graph_its = graph.items()
for it in graph_its:
scene_confed.addItem(it)
it.moveBy(0., header.height())
# add from footer
footer_its = footer.items()
for it in footer_its:
scene_confed.addItem(it)
it.moveBy(0.,header.height() + graph.height())
try:
import OWDlgs
except:
print "Missing file 'OWDlgs.py'. This file should be in OrangeWidgets folder. Unable to print/save image."
sizeDlg = OWDlgs.OWChooseImageSizeDlg(scene_confed, parent=self)
sizeDlg.exec_()
# set all items back to original canvases
for it in header_its:
header.addItem(it)
for it in graph_its:
graph.addItem(it)
it.moveBy(0., -header.height())
for it in footer_its:
footer.addItem(it)
it.moveBy(0, - header.height() - graph.height())
self.showNomogram()
# Callbacks
def showNomogram(self):
if self.bnomogram and self.cl:
#self.bnomogram.hide()
self.bnomogram.show()
self.bnomogram.update()