class RatioMetric(Metric):
numerator = RExpressionTrait()
denominator = RExpressionTrait()
# Whether to apply a logarithm transformation to the ratio.
log_transform = Bool(False)
# Whether to the cap the ratio at certain minimum and maximum values.
cap_below = Bool(False)
cap_below_at = Float(0.0)
cap_below_with = Float(0.0)
cap_above = Bool(False)
cap_above_at = Float(1.0)
cap_above_with = Float(1.0)
# Whether to replace zeros, Infs, and NAs.
replace_zero = Bool(False)
replace_zero_with = Float()
replace_inf = Bool(False)
replace_inf_with = Float()
replace_na = Bool(False)
replace_na_with = Float()
def _ast_impl(self):
args = [ast.Raw(self.numerator), ast.Raw(self.denominator)]
if self.cap_below:
args += [(ast.Name('min'), ast.Constant(self.cap_below_at))]
if self.cap_below_at != self.cap_below_with:
args += [(ast.Name('min_to'),
ast.Constant(self.cap_below_with))]
if self.cap_above:
args += [(ast.Name('max'), ast.Constant(self.cap_above_at))]
if self.cap_above_at != self.cap_above_with:
args += [(ast.Name('max_to'),
ast.Constant(self.cap_above_with))]
if self.replace_zero:
args += [(ast.Name('zero_to'),
ast.Constant(self.replace_zero_with))]
if self.replace_inf:
args += [(ast.Name('inf_to'), ast.Constant(self.replace_inf_with))]
if self.replace_na:
args += [(ast.Name('na_to'), ast.Constant(self.replace_na_with))]
node = ast.Call(ast.Name('ratio'), args)
if self.log_transform:
node = ast.Call(ast.Name('safe_log1p'), node)
return node
def _cap_above_at_changed(self):
self.cap_above_with = self.cap_above_at
def _cap_below_at_changed(self):
self.cap_below_with = self.cap_below_at
class NumericalControl(Control):
expression = RExpressionTrait()
auto_breaks = Bool(True)
num_breaks = Range(low=2, value=10)
breaks = List(Either(Int, Float))
closed_on_left = Bool(True)
labels = List(Str)
def _ast_impl(self):
args = [ast.Raw(self.expression)]
if self.auto_breaks:
arg_breaks = ast.Constant(self.num_breaks)
else:
arg_breaks = ast_macros.seq_to_vector(self.breaks)
args.append((ast.Name('breaks'), arg_breaks))
if self.labels:
args.append(
(ast.Name('labels'), ast_macros.seq_to_vector(self.labels)))
if self.closed_on_left:
# By default, R closes on the right.
args.append((ast.Name('right'), ast.Constant(False)))
return ast.Call(ast.Name('cut'), args, print_hint='long')
def validate(self):
num_breaks = self.num_breaks if self.auto_breaks else len(self.breaks)
if self.labels and len(self.labels) != num_breaks - 1:
msg = "Numerical control variable: incorrect number of labels"
raise ModelError(msg)
class GraphDensityMetric(Metric):
expression = RExpressionTrait()
def ast(self):
return ast.Call(ast.Name('def_group_metric'),
ast.Constant(self.name),
ast.Raw(self.expression),
ast.Name('graph_density'),
print_hint='long')
class UniqueContinuousMetric(Metric):
expression = RExpressionTrait()
def ast(self):
return ast.Call(ast.Name('def_group_metric'),
ast.Constant(self.name),
ast.Raw(self.expression),
ast.Name('uniq_cont'),
print_hint='long')
class EntropyMetric(Metric):
expression = RExpressionTrait()
method = Enum('frequent', 'normal')
def ast(self):
return ast.Call(ast.Name('def_group_metric'),
ast.Constant(self.name),
ast.Raw(self.expression),
ast.Name('entropy_disc'),
(ast.Name('type'), ast.Constant(self.method)),
print_hint='long')
class UniqueDiscreteMetric(Metric):
expression = RExpressionTrait()
method = Enum('distinct', 'frequent')
def ast(self):
return ast.Call(ast.Name('def_group_metric'),
ast.Constant(self.name),
ast.Raw(self.expression),
ast.Name('uniq_disc'),
(ast.Name('type'), ast.Constant(self.method)),
print_hint='long')
class DistributionMetric(Metric):
expression = RExpressionTrait()
kind = Enum('chi_square', 'ks', 'custom')
custom_function = RNameTrait()
def ast(self):
R_funcs = {
'chi_square': 'chisq_test',
'ks': 'ks.stat',
'custom': self.custom_function,
}
return ast.Call(ast.Name('def_group_metric'),
ast.Constant(self.name),
ast.Raw(self.expression),
ast.Name(R_funcs[self.kind]),
print_hint='long')
class FactorControl(Control):
expression = RExpressionTrait()
def _ast_impl(self):
return ast.Raw(self.expression)
class ValueMetric(Metric):
expression = RExpressionTrait()
def _ast_impl(self):
return ast.Raw(self.expression)
class CustomScore(CompositeScore):
expression = RExpressionTrait()
def _ast_impl(self):
return ast.Raw(self.expression)