def testMultipleCallsToEvalInputReceiver(self):
graph = tf.Graph()
features1 = {'apple': tf.constant(1.0), 'banana': tf.constant(2.0)}
labels1 = tf.constant(3.0)
receiver_tensors1 = {'examples': tf.placeholder(tf.string)}
features2 = {'cherry': tf.constant(3.0)}
labels2 = {'alpha': tf.constant(4.0), 'bravo': tf.constant(5.0)}
receiver_tensors2 = {'examples': tf.placeholder(tf.string)}
with graph.as_default():
export.EvalInputReceiver(features=features1,
labels=labels1,
receiver_tensors=receiver_tensors1)
feature_keys_collection_name = encoding.with_suffix(
encoding.FEATURES_COLLECTION, encoding.KEY_SUFFIX)
feature_nodes_collection_name = encoding.with_suffix(
encoding.FEATURES_COLLECTION, encoding.NODE_SUFFIX)
label_keys_collection_name = encoding.with_suffix(
encoding.LABELS_COLLECTION, encoding.KEY_SUFFIX)
label_nodes_collection_name = encoding.with_suffix(
encoding.LABELS_COLLECTION, encoding.NODE_SUFFIX)
self.assertEqual(
2, len(tf.get_collection(feature_keys_collection_name)))
self.assertEqual(
2, len(tf.get_collection(feature_nodes_collection_name)))
self.assertEqual(
1, len(tf.get_collection(label_keys_collection_name)))
self.assertEqual(
1, len(tf.get_collection(label_nodes_collection_name)))
self.assertEqual(
1, len(tf.get_collection(encoding.EXAMPLE_REF_COLLECTION)))
self.assertEqual(
1, len(tf.get_collection(encoding.TFMA_VERSION_COLLECTION)))
# Call again with a different set of features, labels and receiver
# tensors, check that the latest call overrides the earlier one.
#
# Note that we only check the lengths of some collections: more detailed
# checks would require the test to include more knowledge about the
# details of how exporting is done.
export.EvalInputReceiver(features=features2,
labels=labels2,
receiver_tensors=receiver_tensors2)
self.assertEqual(
1, len(tf.get_collection(feature_keys_collection_name)))
self.assertEqual(
1, len(tf.get_collection(feature_nodes_collection_name)))
self.assertEqual(
2, len(tf.get_collection(label_keys_collection_name)))
self.assertEqual(
2, len(tf.get_collection(label_nodes_collection_name)))
self.assertEqual(
1, len(tf.get_collection(encoding.EXAMPLE_REF_COLLECTION)))
self.assertEqual(
1, len(tf.get_collection(encoding.TFMA_VERSION_COLLECTION)))
def eval_input_receiver_fn():
"""Eval input receiver function."""
serialized_tf_example = tf.compat.v1.placeholder(
dtype=tf.string, shape=[None], name='input_example_tensor')
language = tf.feature_column.categorical_column_with_vocabulary_list(
'language', ['english', 'chinese', 'other'])
age = tf.feature_column.numeric_column('age')
english_label = tf.feature_column.numeric_column('english_label')
chinese_label = tf.feature_column.numeric_column('chinese_label')
other_label = tf.feature_column.numeric_column('other_label')
all_features = [
age, language, english_label, chinese_label, other_label
]
feature_spec = tf.feature_column.make_parse_example_spec(all_features)
receiver_tensors = {'examples': serialized_tf_example}
features = tf.io.parse_example(serialized=serialized_tf_example,
features=feature_spec)
labels = {
'english_head': features['english_label'],
'chinese_head': features['chinese_label'],
'other_head': features['other_label'],
}
return export.EvalInputReceiver(features=features,
receiver_tensors=receiver_tensors,
labels=labels)
def eval_input_receiver_fn():
"""Eval input receiver function."""
serialized_tf_example = tf.placeholder(
dtype=tf.string, shape=[None], name='input_example_tensor')
language = tf.contrib.layers.sparse_column_with_keys(
'language', ['english', 'chinese', 'other'])
age = tf.contrib.layers.real_valued_column('age')
english_label = tf.contrib.layers.real_valued_column('english_label')
chinese_label = tf.contrib.layers.real_valued_column('chinese_label')
other_label = tf.contrib.layers.real_valued_column('other_label')
all_features = [age, language, english_label, chinese_label, other_label]
feature_spec = tf.contrib.layers.create_feature_spec_for_parsing(
all_features)
receiver_tensors = {'examples': serialized_tf_example}
features = tf.parse_example(serialized_tf_example, feature_spec)
labels = {
'english_head': features['english_label'],
'chinese_head': features['chinese_label'],
'other_head': features['other_label'],
}
return export.EvalInputReceiver(
features=features, receiver_tensors=receiver_tensors, labels=labels)
def testEvalInputReceiverReceiverTensorKeyCheck(self):
with self.assertRaisesRegexp(ValueError,
'exactly one key named examples'):
export.EvalInputReceiver(
features={},
labels={},
receiver_tensors={'bad_key': tf.constant(0.0)})
def eval_input_receiver_fn():
"""Eval input receiver function."""
csv_row = tf.placeholder(dtype=tf.string,
shape=[None],
name='input_csv_row')
features = parse_csv(csv_row)
receiver_tensors = {'examples': csv_row}
return export.EvalInputReceiver(features=features,
receiver_tensors=receiver_tensors,
labels=features['label'])
def _eval_input_receiver_fn():
"""Eval input receiver function."""
csv_row = tf.placeholder(dtype=tf.string,
shape=[None],
name='input_csv_row')
features = _parse_csv(csv_row)
receiver_tensors = {'examples': csv_row}
# the constructor of EvalInputReceiver() has side effects (populating some
# TF collections). Calling twice here to make sure the collisions are handled
# correctly.
export.EvalInputReceiver(features=features,
labels=features['input_index'],
receiver_tensors=receiver_tensors,
example_ref=features['input_index'])
return export.EvalInputReceiver(features=features,
labels=features['input_index'],
receiver_tensors=receiver_tensors,
example_ref=features['input_index'])
def _bad_eval_input_receiver_fn_out_of_range_input_refs():
"""A bad eval input receiver function (input_refs has out-of-range index)."""
csv_row = tf.compat.v1.placeholder(dtype=tf.string,
shape=[None],
name='input_csv_row')
features = _parse_csv(csv_row)
receiver_tensors = {'examples': csv_row}
return export.EvalInputReceiver(features=features,
labels=features['input_index'],
receiver_tensors=receiver_tensors,
input_refs=features['input_index'] + 1)
def eval_input_receiver_fn():
"""Eval input receiver function."""
serialized_tf_example = tf.placeholder(dtype=tf.string,
shape=[None],
name='input_example_tensor')
receiver_tensors = {'examples': serialized_tf_example}
features = tf.parse_example(serialized_tf_example, label_feature_spec)
_make_embedding_and_sparse_values(features)
return export.EvalInputReceiver(features=features,
receiver_tensors=receiver_tensors,
labels=features['label'])
def eval_input_receiver_fn():
"""An input_fn that expects a serialized tf.Example."""
serialized_tf_example = tf.placeholder(dtype=tf.string,
shape=[None],
name='input_example_tensor')
features = tf.parse_example(serialized_tf_example, eval_feature_spec)
labels = {
'actual_label': features['label'],
'var_int': features['var_int']
}
return export.EvalInputReceiver(
features=features,
labels=labels,
receiver_tensors={'examples': serialized_tf_example})
def _bad_eval_input_receiver_fn_misaligned_input_refs():
"""A bad eval input receiver function capturing a misaligned input_refs."""
csv_row = tf.compat.v1.placeholder(
dtype=tf.string, shape=[None], name='input_csv_row')
features = _parse_csv(csv_row)
receiver_tensors = {'examples': csv_row}
return export.EvalInputReceiver(
features=features,
labels=features['input_index'],
receiver_tensors=receiver_tensors,
input_refs=tf.concat(
[features['input_index'],
tf.constant([0], dtype=tf.int32)], axis=0))
def eval_input_receiver_fn():
"""Eval input receiver function."""
serialized_tf_example = tf.compat.v1.placeholder(
dtype=tf.string, shape=[None], name='input_example_tensor')
feature_spec = {
'age': tf.io.FixedLenFeature([1], dtype=tf.float32),
'label': tf.io.FixedLenFeature([1], dtype=tf.float32)
}
receiver_tensors = {'examples': serialized_tf_example}
features = tf.io.parse_example(serialized=serialized_tf_example,
features=feature_spec)
return export.EvalInputReceiver(features=features,
labels=features['label'],
receiver_tensors=receiver_tensors)
def _eval_input_receiver_using_iterator_fn():
"""Eval input receiver function using an iterator."""
csv_row = tf.compat.v1.placeholder(dtype=tf.string,
shape=[None],
name='input_csv_row')
iterator = tf.compat.v1.data.make_initializable_iterator(
tf.compat.v1.data.Dataset.from_tensors(csv_row))
features = _parse_csv(iterator.get_next())
receiver_tensors = {'examples': csv_row}
return export.EvalInputReceiver(
features=features,
labels=features['input_index'],
receiver_tensors=receiver_tensors,
input_refs=features['input_index'],
iterator_initializer=iterator.initializer.name)
def eval_input_receiver_fn():
"""An input_fn that expects a serialized tf.Example."""
# Note it's *required* that the batch size should be variable for TFMA.
serialized_tf_example = tf.compat.v1.placeholder(
dtype=tf.string, shape=[None], name='input_example_tensor')
features = tf.io.parse_example(serialized=serialized_tf_example,
features=feature_spec)
labels = None if label_key is None else features[label_key]
if isinstance(labels, tf.SparseTensor):
# This bit here is why a custom eval_input_receiver_fn is specified.
labels = tf.sparse.to_dense(labels, default_value=-1)
return export.EvalInputReceiver(
features=features,
labels=labels,
receiver_tensors={'examples': serialized_tf_example})
def eval_input_receiver_fn():
"""Eval input receiver function."""
serialized_tf_example = tf.placeholder(
dtype=tf.string, shape=[None], name='input_example_tensor')
animals = tf.contrib.layers.sparse_column_with_keys('animals',
['bird', 'cat', 'dog'])
label = tf.contrib.layers.real_valued_column('label')
all_features = [animals, label]
feature_spec = tf.contrib.layers.create_feature_spec_for_parsing(
all_features)
receiver_tensors = {'examples': serialized_tf_example}
features = tf.parse_example(serialized_tf_example, feature_spec)
return export.EvalInputReceiver(
features=features,
receiver_tensors=receiver_tensors,
labels=features['label'])
def eval_input_receiver_fn():
"""Eval input receiver function."""
serialized_tf_example = tf.placeholder(dtype=tf.string,
shape=[None],
name='input_example_tensor')
feature_spec = {
'prediction': tf.FixedLenFeature([1], dtype=tf.float32),
'label': tf.FixedLenFeature([1], dtype=tf.float32),
'fixed_float': tf.FixedLenFeature([1], dtype=tf.float32),
'fixed_string': tf.FixedLenFeature([1], dtype=tf.string),
'var_float': tf.VarLenFeature(dtype=tf.float32),
'var_string': tf.VarLenFeature(dtype=tf.string)
}
receiver_tensors = {'examples': serialized_tf_example}
features = tf.parse_example(serialized_tf_example, feature_spec)
return export.EvalInputReceiver(features=features,
receiver_tensors=receiver_tensors,
labels=features['label'])
def eval_input_receiver_fn():
"""Eval input receiver function."""
serialized_tf_example = tf.placeholder(dtype=tf.string,
shape=[None],
name='input_example_tensor')
language = tf.contrib.layers.sparse_column_with_keys(
'language', ['english', 'chinese'])
slice_key = tf.contrib.layers.sparse_column_with_hash_bucket(
'slice_key', 100)
age = tf.contrib.layers.real_valued_column('age')
label = tf.contrib.layers.real_valued_column('label')
all_features = [age, language, label, slice_key]
feature_spec = tf.contrib.layers.create_feature_spec_for_parsing(
all_features)
receiver_tensors = {'examples': serialized_tf_example}
features = tf.parse_example(serialized_tf_example, feature_spec)
return export.EvalInputReceiver(features=features,
receiver_tensors=receiver_tensors,
labels=features['label'])