def test_get_categorical_features_to_sampling(self):
cat_example = tf.train.Example()
cat_example.features.feature['non_numeric'].bytes_list.value.extend(
[b'cat'])
cow_example = tf.train.Example()
cow_example.features.feature['non_numeric'].bytes_list.value.extend(
[b'cow'])
pony_example = tf.train.Example()
pony_example.features.feature['non_numeric'].bytes_list.value.extend(
[b'pony'])
examples = [cat_example] * 4 + [cow_example] * 5 + [pony_example] * 10
# If we stop sampling at the first 3 examples, the only example should be
# cat example.
data = inference_utils.get_categorical_features_to_sampling(
examples[0:3], top_k=1)
self.assertDictEqual({'non_numeric': {'samples': [b'cat']}}, data)
# If we sample more examples, the top 2 examples should be cow and pony.
data = inference_utils.get_categorical_features_to_sampling(
examples[0:20], top_k=2)
self.assertDictEqual({'non_numeric': {
'samples': [b'pony', b'cow']
}}, data)
def _eligible_features_from_example_handler(self, request):
"""Returns a list of JSON objects for each feature in the example.
Args:
request: A request for features.
Returns:
A list with a JSON object for each feature.
Numeric features are represented as {name: observedMin: observedMax:}.
Categorical features are repesented as {name: samples:[]}.
"""
features_dict = (
inference_utils.get_numeric_features_to_observed_range(
self.examples[0:NUM_EXAMPLES_TO_SCAN]))
features_dict.update(
inference_utils.get_categorical_features_to_sampling(
self.examples[0:NUM_EXAMPLES_TO_SCAN], NUM_MUTANTS))
# Massage the features_dict into a sorted list before returning because
# Polymer dom-repeat needs a list.
features_list = []
for k, v in sorted(features_dict.items()):
v['name'] = k
features_list.append(v)
return http_util.Respond(request, features_list, 'application/json')
def _eligible_features_from_example_handler(self, request):
"""Returns a list of JSON objects for each feature in the example.
Args:
request: A request for features.
Returns:
A list with a JSON object for each feature.
Numeric features are represented as {name: observedMin: observedMax:}.
Categorical features are repesented as {name: samples:[]}.
"""
features_dict = (
inference_utils.get_numeric_features_to_observed_range(
self.examples[0: NUM_EXAMPLES_TO_SCAN]))
features_dict.update(
inference_utils.get_categorical_features_to_sampling(
self.examples[0: NUM_EXAMPLES_TO_SCAN], NUM_MUTANTS))
# Massage the features_dict into a sorted list before returning because
# Polymer dom-repeat needs a list.
features_list = []
for k, v in sorted(features_dict.items()):
v['name'] = k
features_list.append(v)
return http_util.Respond(request, features_list, 'application/json')
