def testMemoizedInferrerLoading(self):
inferrer = inference.memoized_inferrer(test_util.savedmodel_path(),
memoize_inference_results=True)
memoized_inferrer = inference.memoized_inferrer(
test_util.savedmodel_path(), memoize_inference_results=True)
self.assertIs(inferrer, memoized_inferrer)
def testMemoizedInferenceResults(self):
inferrer = inference.Inferrer(test_util.savedmodel_path(),
memoize_inference_results=True)
activations = inferrer._get_activations_for_batch(('ADE', ))
memoized_activations = inferrer._get_activations_for_batch(('ADE', ))
self.assertIs(activations, memoized_activations)
def testStringInput(self):
inferrer = inference.Inferrer(test_util.savedmodel_path())
# Simulate failure to use a list.
with self.assertRaisesRegex(
ValueError, '`list_of_seqs` should be convertible to a '
'numpy vector of strings. Got *'):
inferrer.get_activations('QP')
def testBatchedInference(self):
inferrer = inference.Inferrer(test_util.savedmodel_path(),
batch_size=5)
input_seq = 'AP'
for total_size in range(15):
full_list = [input_seq] * total_size
activations = inferrer.get_activations(full_list)
self.assertLen(full_list, activations.shape[0])
def testSortUnsortInference(self):
inferrer = inference.Inferrer(test_util.savedmodel_path(),
batch_size=1)
input_seqs = ['AP', 'APP', 'AP']
# Sorting will move long sequence to the end.
activations = inferrer.get_activations(input_seqs)
# Make sure it gets moved back to the middle.
self.assertAllClose(activations[0], activations[2])
self.assertNotAllClose(activations[0], activations[1])
def testCanInfer(self):
graph = tf.Graph()
with graph.as_default():
sequences = tf.placeholder(shape=[None], dtype=tf.string)
output_tensor = inference.in_graph_inferrer(
sequences, test_util.savedmodel_path(), tf.saved_model.
signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY)
input_seqs = [''.join(utils.FULL_RESIDUE_VOCAB), 'ACD']
with self.session(graph=graph) as sess:
sess.run(tf.global_variables_initializer())
sess.run(tf.tables_initializer())
result = sess.run(output_tensor, feed_dict={sequences: input_seqs})
self.assertLen(result, 2)
def testGetVariable(self):
inferrer = inference.Inferrer(test_util.savedmodel_path())
output = inferrer.get_variable('conv1d/bias:0')
self.assertNotEmpty(output)