def testConvertEvalResultToUIInputWithSlicingColumn(self):
eval_result = self._makeEvalResult()
result = widget_view.convert_slicing_metrics_to_ui_input(
eval_result.slicing_metrics, slicing_column='slice')
self.assertEqual(result, [{
'slice': 'Overall',
'sliceValue': 'Overall',
'metrics': {
'metrics2': {
'double_value': {
'value': 0.5
}
},
'metrics1': {
'double_value': {
'value': 0.5
}
}
}
}, {
'slice': u'slice:1',
'sliceValue': u'1',
'metrics': {
'metrics2': {
'double_value': {
'value': 0.5
}
},
'metrics1': {
'double_value': {
'value': 0.5
}
}
}
}])
def _get_evaluation_result_from_remote_path(self, request):
evaluation_output_path = request.args.get('evaluation_output_path')
try:
evaluation_output_path = six.ensure_text(evaluation_output_path)
except (UnicodeDecodeError, AttributeError):
pass
try:
metrics = (
metrics_and_plots_serialization.load_and_deserialize_metrics(
path=evaluation_output_path))
data = widget_view.convert_slicing_metrics_to_ui_input(metrics)
except (KeyError, json_format.ParseError) as error:
logging.info('Error while fetching evaluation data, %s', error)
data = []
return http_util.Respond(request,
data,
content_type='application/json')
def _get_evaluation_result_from_remote_path(self, request):
evaluation_output_path = request.args.get('evaluation_output_path')
try:
evaluation_output_path = six.ensure_text(evaluation_output_path)
except (UnicodeDecodeError, AttributeError):
pass
try:
eval_result = tfma.load_eval_result(
os.path.dirname(evaluation_output_path),
output_file_format=self._get_output_file_format(
evaluation_output_path))
data = widget_view.convert_slicing_metrics_to_ui_input(
eval_result.slicing_metrics)
except (KeyError, json_format.ParseError) as error:
logging.info('Error while fetching evaluation data, %s', error)
data = []
return http_util.Respond(request,
data,
content_type='application/json')
def testConvertEvalResultToUIInputForCrossSliceKeyType(self):
eval_result = self._makeEvalResult(slices=(((), (('slice_1', 1),
('slice_2', 2))),
((('slice_1', 5),
('slice_2', 6)),
(('slice_1', 3),
('slice_2', 4)))))
result = widget_view.convert_slicing_metrics_to_ui_input(
eval_result.slicing_metrics)
self.assertEqual(result, [{
'slice': 'Overall__XX__slice_1_X_slice_2:1_X_2',
'sliceValue': 'Overall__XX__1_X_2',
'metrics': {
'metrics2': {
'double_value': {
'value': 0.5
}
},
'metrics1': {
'double_value': {
'value': 0.5
}
}
}
}, {
'slice': 'slice_1_X_slice_2:5_X_6__XX__slice_1_X_slice_2:3_X_4',
'sliceValue': '5_X_6__XX__3_X_4',
'metrics': {
'metrics2': {
'double_value': {
'value': 0.5
}
},
'metrics1': {
'double_value': {
'value': 0.5
}
}
}
}])
def _get_evaluation_result(self, request):
run = request.args.get('run')
try:
run = six.ensure_text(run)
except (UnicodeDecodeError, AttributeError):
pass
data = []
try:
eval_result_output_dir = six.ensure_text(
self._multiplexer.Tensors(run,
FairnessIndicatorsPlugin.plugin_name)
[0].tensor_proto.string_val[0])
eval_result = tfma.load_eval_result(
output_path=eval_result_output_dir)
# TODO(b/141283811): Allow users to choose different model output names
# and class keys in case of multi-output and multi-class model.
data = widget_view.convert_slicing_metrics_to_ui_input(
eval_result.slicing_metrics)
except (KeyError, json_format.ParseError) as error:
logging.info('Error while fetching evaluation data, %s', error)
return http_util.Respond(request,
data,
content_type='application/json')
