def testEvaluateQueryBasedMetrics(self):
temp_eval_export_dir = self._getEvalExportDir()
_, eval_export_dir = (fixed_prediction_estimator_extra_fields.
simple_fixed_prediction_estimator_extra_fields(
None, temp_eval_export_dir))
eval_shared_model = self.createTestEvalSharedModel(
eval_saved_model_path=eval_export_dir)
extractors = [
legacy_predict_extractor.PredictExtractor(eval_shared_model),
slice_key_extractor.SliceKeyExtractor()
]
with beam.Pipeline() as pipeline:
metrics = (
pipeline
| 'Create' >> beam.Create(self._get_examples())
| 'InputsToExtracts' >> model_eval_lib.InputsToExtracts()
| 'Extract' >> tfma_unit.Extract(extractors=extractors) # pylint: disable=no-value-for-parameter
| 'EvaluateQueryBasedMetrics' >>
query_based_metrics_evaluator.EvaluateQueryBasedMetrics(
prediction_key='',
query_id='fixed_string',
combine_fns=[
query_statistics.QueryStatisticsCombineFn(),
ndcg.NdcgMetricCombineFn(at_vals=[1, 2],
gain_key='fixed_float',
weight_key='fixed_int'),
min_label_position.MinLabelPositionCombineFn(
label_key='', weight_key='fixed_int'),
]))
def check_metrics(got):
try:
self.assertEqual(1, len(got), 'got: %s' % got)
got_slice_key, got_metrics = got[0]
self.assertEqual(got_slice_key, ())
self.assertDictElementsAlmostEqual(
got_metrics, {
'post_export_metrics/total_queries':
3.0,
'post_export_metrics/total_documents':
6.0,
'post_export_metrics/min_documents':
1.0,
'post_export_metrics/max_documents':
3.0,
'post_export_metrics/ndcg@1':
0.9166667,
'post_export_metrics/ndcg@2':
0.9766198,
'post_export_metrics/average_min_label_position/__labels':
0.6666667,
})
except AssertionError as err:
raise util.BeamAssertException(err)
util.assert_that(metrics[constants.METRICS_KEY],
check_metrics,
label='metrics')
def testEvaluateNoSlicing(self):
temp_eval_export_dir = self._getEvalExportDir()
_, eval_export_dir = linear_classifier.simple_linear_classifier(
None, temp_eval_export_dir)
eval_shared_model = self.createTestEvalSharedModel(
eval_saved_model_path=eval_export_dir,
add_metrics_callbacks=[_addExampleCountMetricCallback])
extractors = [
predict_extractor.PredictExtractor(eval_shared_model),
slice_key_extractor.SliceKeyExtractor()
]
with beam.Pipeline() as pipeline:
example1 = self._makeExample(age=3.0,
language='english',
label=1.0)
example2 = self._makeExample(age=3.0,
language='chinese',
label=0.0)
example3 = self._makeExample(age=4.0,
language='english',
label=1.0)
example4 = self._makeExample(age=5.0,
language='chinese',
label=0.0)
metrics, _ = (
pipeline
| 'Create' >> beam.Create([
example1.SerializeToString(),
example2.SerializeToString(),
example3.SerializeToString(),
example4.SerializeToString()
])
| 'InputsToExtracts' >> model_eval_lib.InputsToExtracts()
| 'Extract' >> tfma_unit.Extract(extractors=extractors) # pylint: disable=no-value-for-parameter
| 'ComputeMetricsAndPlots' >> metrics_and_plots_evaluator.
ComputeMetricsAndPlots(eval_shared_model=eval_shared_model))
def check_result(got):
try:
self.assertEqual(1, len(got), 'got: %s' % got)
(slice_key, value) = got[0]
self.assertEqual((), slice_key)
self.assertDictElementsAlmostEqual(
value, {
'accuracy': 1.0,
'label/mean': 0.5,
'my_mean_age': 3.75,
'my_mean_age_times_label': 1.75,
'added_example_count': 4.0
})
except AssertionError as err:
raise util.BeamAssertException(err)
util.assert_that(metrics, check_result)
def testEvaluateNoSlicingAddPostExportAndCustomMetricsUnsupervisedModel(
self):
# Mainly for testing that the ExampleCount post export metric works with
# unsupervised models.
temp_eval_export_dir = self._getEvalExportDir()
_, eval_export_dir = (fixed_prediction_estimator_no_labels.
simple_fixed_prediction_estimator_no_labels(
None, temp_eval_export_dir))
eval_shared_model = self.createTestEvalSharedModel(
eval_saved_model_path=eval_export_dir,
add_metrics_callbacks=[
post_export_metrics.example_count(),
post_export_metrics.example_weight(
example_weight_key='prediction')
])
extractors = [
predict_extractor.PredictExtractor(eval_shared_model),
slice_key_extractor.SliceKeyExtractor()
]
with beam.Pipeline() as pipeline:
example1 = self._makeExample(prediction=1.0)
example2 = self._makeExample(prediction=2.0)
metrics, plots = (
pipeline
| 'Create' >> beam.Create([
example1.SerializeToString(),
example2.SerializeToString(),
])
| 'InputsToExtracts' >> model_eval_lib.InputsToExtracts()
| 'Extract' >> tfma_unit.Extract(extractors=extractors) # pylint: disable=no-value-for-parameter
| 'ComputeMetricsAndPlots' >> metrics_and_plots_evaluator.
ComputeMetricsAndPlots(eval_shared_model=eval_shared_model))
def check_result(got):
try:
self.assertEqual(1, len(got), 'got: %s' % got)
(slice_key, value) = got[0]
self.assertEqual((), slice_key)
self.assertDictElementsAlmostEqual(
got_values_dict=value,
expected_values_dict={
'average_loss': 2.5,
metric_keys.EXAMPLE_COUNT: 2.0,
metric_keys.EXAMPLE_WEIGHT: 3.0
})
except AssertionError as err:
raise util.BeamAssertException(err)
util.assert_that(metrics, check_result, label='metrics')
util.assert_that(plots, util.is_empty(), label='plots')
def _runTestWithCustomCheck(self,
examples,
eval_export_dir,
metrics_callbacks,
slice_spec=None,
custom_metrics_check=None,
custom_plots_check=None,
custom_result_check=None):
# make sure we are doing some checks
self.assertTrue(custom_metrics_check is not None
or custom_plots_check is not None
or custom_result_check is not None)
serialized_examples = [ex.SerializeToString() for ex in examples]
slicing_specs = None
if slice_spec:
slicing_specs = [s.to_proto() for s in slice_spec]
eval_config = config.EvalConfig(slicing_specs=slicing_specs)
eval_shared_model = self.createTestEvalSharedModel(
eval_saved_model_path=eval_export_dir,
add_metrics_callbacks=metrics_callbacks)
extractors = model_eval_lib.default_extractors(
eval_config=eval_config, eval_shared_model=eval_shared_model)
tfx_io = raw_tf_record.RawBeamRecordTFXIO(
physical_format='inmemory',
raw_record_column_name=constants.ARROW_INPUT_COLUMN,
telemetry_descriptors=['TFMATest'])
with beam.Pipeline() as pipeline:
(metrics, plots), _ = (
pipeline
| 'Create' >> beam.Create(serialized_examples)
| 'BatchExamples' >> tfx_io.BeamSource()
|
'InputsToExtracts' >> model_eval_lib.BatchedInputsToExtracts()
| 'Extract' >> tfma_unit.Extract(extractors=extractors) # pylint: disable=no-value-for-parameter
| 'ComputeMetricsAndPlots' >>
legacy_metrics_and_plots_evaluator._ComputeMetricsAndPlots( # pylint: disable=protected-access
eval_shared_model=eval_shared_model,
compute_confidence_intervals=self.
compute_confidence_intervals,
random_seed_for_testing=self.deterministic_test_seed))
if custom_metrics_check is not None:
util.assert_that(metrics,
custom_metrics_check,
label='metrics')
if custom_plots_check is not None:
util.assert_that(plots, custom_plots_check, label='plot')
result = pipeline.run()
if custom_result_check is not None:
custom_result_check(result)
def _runTestWithCustomCheck(self,
examples,
eval_export_dir,
metrics_callbacks,
slice_spec=None,
custom_metrics_check=None,
custom_plots_check=None,
custom_result_check=None):
# make sure we are doing some checks
self.assertTrue(custom_metrics_check is not None
or custom_plots_check is not None
or custom_result_check is not None)
serialized_examples = [ex.SerializeToString() for ex in examples]
slicing_specs = None
if slice_spec:
slicing_specs = [s.to_proto() for s in slice_spec]
eval_config = config.EvalConfig(
input_data_specs=[config.InputDataSpec()],
model_specs=[config.ModelSpec(location=eval_export_dir)],
output_data_specs=[config.OutputDataSpec()],
slicing_specs=slicing_specs)
eval_shared_model = self.createTestEvalSharedModel(
eval_saved_model_path=eval_export_dir,
add_metrics_callbacks=metrics_callbacks)
extractors = model_eval_lib.default_extractors(
eval_config=eval_config, eval_shared_models=[eval_shared_model])
with beam.Pipeline() as pipeline:
(metrics, plots), _ = (
pipeline
| 'Create' >> beam.Create(serialized_examples)
| 'InputsToExtracts' >> model_eval_lib.InputsToExtracts()
| 'Extract' >> tfma_unit.Extract(extractors=extractors) # pylint: disable=no-value-for-parameter
| 'ComputeMetricsAndPlots' >>
metrics_and_plots_evaluator.ComputeMetricsAndPlots(
eval_shared_model=eval_shared_model,
compute_confidence_intervals=self.
compute_confidence_intervals,
random_seed_for_testing=self.deterministic_test_seed))
if custom_metrics_check is not None:
util.assert_that(metrics,
custom_metrics_check,
label='metrics')
if custom_plots_check is not None:
util.assert_that(plots, custom_plots_check, label='plot')
result = pipeline.run()
if custom_result_check is not None:
custom_result_check(result)
def testModelAgnosticConstructFn(self):
# End to end test for the entire flow going from tf.Examples -> metrics
# with slicing.
with beam.Pipeline() as pipeline:
# Set up the inputs. All we need is are tf.Examples and an example parsing
# spec with explicit mapping for key to (Features, Predictions, Labels).
examples = [
self._makeExample(age=3.0,
language='english',
probabilities=1.0,
labels=1.0),
self._makeExample(age=3.0,
language='chinese',
probabilities=3.0,
labels=0.0),
self._makeExample(age=4.0,
language='english',
probabilities=2.0,
labels=1.0),
self._makeExample(age=5.0,
language='chinese',
probabilities=3.0,
labels=0.0),
# Add some examples with no language.
self._makeExample(age=5.0, probabilities=2.0, labels=10.0),
self._makeExample(age=6.0, probabilities=1.0, labels=0.0)
]
serialized_examples = [e.SerializeToString() for e in examples]
# Set up a config to bucket our example keys.
feature_map = {
'age': tf.FixedLenFeature([], tf.float32),
'language': tf.VarLenFeature(tf.string),
'probabilities': tf.FixedLenFeature([], tf.float32),
'labels': tf.FixedLenFeature([], tf.float32)
}
model_agnostic_config = agnostic_predict.ModelAgnosticConfig(
label_keys=['labels'],
prediction_keys=['probabilities'],
feature_spec=feature_map)
# Set up the Model Agnostic Extractor
extractors = [
model_agnostic_extractor.ModelAgnosticExtractor(
model_agnostic_config=model_agnostic_config,
desired_batch_size=3),
slice_key_extractor.SliceKeyExtractor([
slicer.SingleSliceSpec(),
slicer.SingleSliceSpec(columns=['language'])
])
]
# Set up the metrics we wish to calculate via a metric callback. In
# particular, this metric calculates the mean and sum of all labels.
eval_shared_model = types.EvalSharedModel(
add_metrics_callbacks=[add_mean_callback],
construct_fn=model_agnostic_evaluate_graph.make_construct_fn(
add_metrics_callbacks=[add_mean_callback],
fpl_feed_config=model_agnostic_extractor.
ModelAgnosticGetFPLFeedConfig(model_agnostic_config)))
# Run our pipeline doing Extract -> Slice -> Fanout -> Calculate Metrics.
metrics, _ = (
pipeline
| 'Create Examples' >> beam.Create(serialized_examples)
| 'InputsToExtracts' >> model_eval_lib.InputsToExtracts()
| 'Extract' >> tfma_unit.Extract(extractors=extractors) # pylint: disable=no-value-for-parameter
| 'ComputeMetricsAndPlots' >> metrics_and_plots_evaluator.
ComputeMetricsAndPlots(eval_shared_model=eval_shared_model))
# Verify our metrics are properly generated per slice.
def check_result(got):
self.assertEqual(3, len(got), 'got: %s' % got)
slices = {}
for slice_key, metrics in got:
slices[slice_key] = metrics
overall_slice = ()
english_slice = (('language', b'english'), )
chinese_slice = (('language', b'chinese'), )
self.assertItemsEqual(
list(slices.keys()),
[overall_slice, english_slice, chinese_slice])
# Overall slice has label/predictions sum = 24 and 12 elements.
self.assertDictElementsAlmostEqual(slices[overall_slice], {
'tf_metric_mean': 2.0,
'py_func_total_label': 24.0,
})
# English slice has label/predictions sum = 5 and 4 elements.
self.assertDictElementsAlmostEqual(slices[english_slice], {
'tf_metric_mean': 1.25,
'py_func_total_label': 5.0,
})
# Chinese slice has label/predictions sum = 6 and 4 elements.
self.assertDictElementsAlmostEqual(slices[chinese_slice], {
'tf_metric_mean': 1.5,
'py_func_total_label': 6.0,
})
util.assert_that(metrics, check_result)
def testEvaluateWithPlots(self):
temp_eval_export_dir = self._getEvalExportDir()
_, eval_export_dir = (
fixed_prediction_estimator.simple_fixed_prediction_estimator(
None, temp_eval_export_dir))
eval_shared_model = self.createTestEvalSharedModel(
eval_saved_model_path=eval_export_dir,
add_metrics_callbacks=[
post_export_metrics.example_count(),
post_export_metrics.auc_plots()
])
extractors = [
predict_extractor.PredictExtractor(eval_shared_model),
slice_key_extractor.SliceKeyExtractor()
]
with beam.Pipeline() as pipeline:
example1 = self._makeExample(prediction=0.0, label=1.0)
example2 = self._makeExample(prediction=0.7, label=0.0)
example3 = self._makeExample(prediction=0.8, label=1.0)
example4 = self._makeExample(prediction=1.0, label=1.0)
metrics, plots = (
pipeline
| 'Create' >> beam.Create([
example1.SerializeToString(),
example2.SerializeToString(),
example3.SerializeToString(),
example4.SerializeToString()
])
| 'InputsToExtracts' >> model_eval_lib.InputsToExtracts()
| 'Extract' >> tfma_unit.Extract(extractors=extractors) # pylint: disable=no-value-for-parameter
| 'ComputeMetricsAndPlots' >> metrics_and_plots_evaluator.
ComputeMetricsAndPlots(eval_shared_model=eval_shared_model))
def check_metrics(got):
try:
self.assertEqual(1, len(got), 'got: %s' % got)
(slice_key, value) = got[0]
self.assertEqual((), slice_key)
self.assertDictElementsAlmostEqual(
got_values_dict=value,
expected_values_dict={
metric_keys.EXAMPLE_COUNT: 4.0,
})
except AssertionError as err:
raise util.BeamAssertException(err)
util.assert_that(metrics, check_metrics, label='metrics')
def check_plots(got):
try:
self.assertEqual(1, len(got), 'got: %s' % got)
(slice_key, value) = got[0]
self.assertEqual((), slice_key)
self.assertDictMatrixRowsAlmostEqual(
got_values_dict=value,
expected_values_dict={
_full_key(metric_keys.AUC_PLOTS_MATRICES):
[(8001, [2, 1, 0, 1, 1.0 / 1.0, 1.0 / 3.0])],
})
except AssertionError as err:
raise util.BeamAssertException(err)
util.assert_that(plots, check_plots, label='plots')
def testEvaluateNoSlicingAddPostExportAndCustomMetrics(self):
temp_eval_export_dir = self._getEvalExportDir()
_, eval_export_dir = linear_classifier.simple_linear_classifier(
None, temp_eval_export_dir)
eval_shared_model = self.createTestEvalSharedModel(
eval_saved_model_path=eval_export_dir,
add_metrics_callbacks=[
_addExampleCountMetricCallback,
# Note that since everything runs in-process this doesn't
# actually test that the py_func can be correctly recreated
# on workers in a distributed context.
_addPyFuncMetricCallback,
post_export_metrics.example_count(),
post_export_metrics.example_weight(example_weight_key='age')
])
extractors = [
predict_extractor.PredictExtractor(eval_shared_model),
slice_key_extractor.SliceKeyExtractor()
]
with beam.Pipeline() as pipeline:
example1 = self._makeExample(age=3.0,
language='english',
label=1.0)
example2 = self._makeExample(age=3.0,
language='chinese',
label=0.0)
example3 = self._makeExample(age=4.0,
language='english',
label=1.0)
example4 = self._makeExample(age=5.0,
language='chinese',
label=0.0)
metrics, plots = (
pipeline
| 'Create' >> beam.Create([
example1.SerializeToString(),
example2.SerializeToString(),
example3.SerializeToString(),
example4.SerializeToString()
])
| 'InputsToExtracts' >> model_eval_lib.InputsToExtracts()
| 'Extract' >> tfma_unit.Extract(extractors=extractors) # pylint: disable=no-value-for-parameter
| 'ComputeMetricsAndPlots' >> metrics_and_plots_evaluator.
ComputeMetricsAndPlots(eval_shared_model=eval_shared_model))
def check_result(got):
try:
self.assertEqual(1, len(got), 'got: %s' % got)
(slice_key, value) = got[0]
self.assertEqual((), slice_key)
self.assertDictElementsAlmostEqual(
got_values_dict=value,
expected_values_dict={
'accuracy': 1.0,
'label/mean': 0.5,
'my_mean_age': 3.75,
'my_mean_age_times_label': 1.75,
'added_example_count': 4.0,
'py_func_label_sum': 2.0,
metric_keys.EXAMPLE_COUNT: 4.0,
metric_keys.EXAMPLE_WEIGHT: 15.0
})
except AssertionError as err:
raise util.BeamAssertException(err)
util.assert_that(metrics, check_result, label='metrics')
util.assert_that(plots, util.is_empty(), label='plots')
def testEvaluateWithSlicingAndDifferentBatchSizes(self):
temp_eval_export_dir = self._getEvalExportDir()
_, eval_export_dir = linear_classifier.simple_linear_classifier(
None, temp_eval_export_dir)
eval_shared_model = self.createTestEvalSharedModel(
eval_saved_model_path=eval_export_dir,
add_metrics_callbacks=[_addExampleCountMetricCallback])
extractors = [
predict_extractor.PredictExtractor(eval_shared_model),
slice_key_extractor.SliceKeyExtractor([
slicer.SingleSliceSpec(),
slicer.SingleSliceSpec(columns=['slice_key'])
])
]
for batch_size in [1, 2, 4, 8]:
with beam.Pipeline() as pipeline:
example1 = self._makeExample(age=3.0,
language='english',
label=1.0,
slice_key='first_slice')
example2 = self._makeExample(age=3.0,
language='chinese',
label=0.0,
slice_key='first_slice')
example3 = self._makeExample(age=4.0,
language='english',
label=0.0,
slice_key='second_slice')
example4 = self._makeExample(age=5.0,
language='chinese',
label=1.0,
slice_key='second_slice')
example5 = self._makeExample(age=5.0,
language='chinese',
label=1.0,
slice_key='second_slice')
metrics, plots = (
pipeline
| 'Create' >> beam.Create([
example1.SerializeToString(),
example2.SerializeToString(),
example3.SerializeToString(),
example4.SerializeToString(),
example5.SerializeToString(),
])
| 'InputsToExtracts' >> model_eval_lib.InputsToExtracts()
| 'Extract' >> tfma_unit.Extract(extractors=extractors) # pylint: disable=no-value-for-parameter
| 'ComputeMetricsAndPlots' >>
metrics_and_plots_evaluator.ComputeMetricsAndPlots(
eval_shared_model=eval_shared_model,
desired_batch_size=batch_size))
def check_result(got):
try:
self.assertEqual(3, len(got), 'got: %s' % got)
slices = {}
for slice_key, value in got:
slices[slice_key] = value
overall_slice = ()
first_slice = (('slice_key', b'first_slice'), )
second_slice = (('slice_key', b'second_slice'), )
self.assertItemsEqual(
list(slices.keys()),
[overall_slice, first_slice, second_slice])
self.assertDictElementsAlmostEqual(
slices[overall_slice], {
'accuracy': 0.4,
'label/mean': 0.6,
'my_mean_age': 4.0,
'my_mean_age_times_label': 2.6,
'added_example_count': 5.0
})
self.assertDictElementsAlmostEqual(
slices[first_slice], {
'accuracy': 1.0,
'label/mean': 0.5,
'my_mean_age': 3.0,
'my_mean_age_times_label': 1.5,
'added_example_count': 2.0
})
self.assertDictElementsAlmostEqual(
slices[second_slice], {
'accuracy': 0.0,
'label/mean': 2.0 / 3.0,
'my_mean_age': 14.0 / 3.0,
'my_mean_age_times_label': 10.0 / 3.0,
'added_example_count': 3.0
})
except AssertionError as err:
# This function is redefined every iteration, so it will have the
# right value of batch_size.
raise util.BeamAssertException(
'batch_size = %d, error: %s' % (batch_size, err)) # pylint: disable=cell-var-from-loop
util.assert_that(metrics, check_result, label='metrics')
util.assert_that(plots, util.is_empty(), label='plots')
