def plot(self):
self.pimp.table_for_comparison(self.evaluators, self.out_fn, style='latex')
self.logger.info('Creating pimp latex table at %s' % self.out_fn)
parameters = [p.name for p in self.cs.get_hyperparameters()]
index, values, columns = [], [], []
columns = [e.name for e in self.evaluators]
columns_lower = [c.lower() for c in columns]
# SORT
self.logger.debug("Sort pimp-table by %s" % self.sort_table_by)
if self.sort_table_by == "average":
# Sort parameters after average importance
p_avg = {}
for p in parameters:
imps = [e.evaluated_parameter_importance[p] for e in self.evaluators if p in e.evaluated_parameter_importance]
p_avg[p] = np.mean(imps) if imps else 0
p_order = sorted(parameters, key=lambda p: p_avg[p], reverse=True)
elif self.sort_table_by in columns_lower:
def __get_key(p):
imp = self.evaluators[columns_lower.index(self.sort_table_by)].evaluated_parameter_importance
return imp[p] if p in imp else 0
p_order = sorted(parameters, key=__get_key, reverse=True)
else:
raise ValueError("Trying to sort importance table after {}, which "
"was not evaluated.".format(self.sort_table_by))
# PREPROCESS
for p in p_order:
values_for_p = [p]
add_parameter = False # Only add parameters where at least one evaluator shows importance > threshold
for e in self.evaluators:
if p in e.evaluated_parameter_importance:
# Check for threshold
value_to_add = e.evaluated_parameter_importance[p]
if value_to_add > self.threshold:
add_parameter = True
# All but forward-selection use values between 0 and 1
if e.name != 'Forward-Selection':
value_to_add = value_to_add * 100
# Create string and add uncertainty, if available
value_to_add = format(value_to_add, '05.2f') # (leading zeros for sorting!)
if (hasattr(e, 'evaluated_parameter_importance_uncertainty') and
p in e.evaluated_parameter_importance_uncertainty):
value_to_add += ' +/- ' + format(e.evaluated_parameter_importance_uncertainty[p] * 100, '.2f')
values_for_p.append(value_to_add)
else:
values_for_p.append('-')
if add_parameter:
values.append(values_for_p)
# CREATE TABLE
self.comp_table = DataFrame(values, columns=['Parameters'] + columns)
sortable = {c : True for c in columns}
width = {**{'Parameters' : 150}, **{c : 100 for c in columns}}
bokeh_table = array_to_bokeh_table(self.comp_table, sortable=sortable, width=width, logger=self.logger)
return bokeh_table
def test_array_to_bokeh_table(self):
dataframe = DataFrame(self.rng.rand(2, 3),
columns=[str(i) for i in range(3)])
self.assertTrue(isinstance(array_to_bokeh_table(dataframe), DataTable))
# Pass logger
self.assertTrue(
isinstance(
array_to_bokeh_table(dataframe,
logger=logging.getLogger('test')),
DataTable))
# Pass sortable and width
self.assertTrue(
isinstance(
array_to_bokeh_table(dataframe,
sortable={
'1': True,
'2': True
},
width={
'1': 100,
'0': 200
}), DataTable))
# Pass invalid specifications
self.assertRaises(ValueError,
array_to_bokeh_table,
dataframe,
sortable={
'7': True,
'2': True
})
self.assertRaises(ValueError,
array_to_bokeh_table,
dataframe,
width={
'1': 100,
10: 200
})