def testCosineSimilarityProviderMatching(self):
"""Check correctness for querying the library with a library element."""
num_examples = 20
num_trials = 10
data_dim = 5
similarity = similarity_lib.CosineSimilarityProvider()
library = np.float32(np.random.normal(size=(num_examples, data_dim)))
library = tf.constant(library)
library = similarity.preprocess_library(library)
query_idx = tf.placeholder(shape=(), dtype=tf.int32)
query = library[query_idx][np.newaxis, ...]
(match_idx_op, match_similarity_op, _, _,
_) = library_matching._max_similarity_match(library, query,
similarity)
# Use queries that are rows of the library. This means that the maximum
# cosine similarity is 1.0 and is achieved by the row index of the query
# in the library.
with tf.Session() as sess:
for _ in range(num_trials):
idx = np.random.randint(0, high=num_examples)
match_idx, match_similarity = sess.run(
[match_idx_op, match_similarity_op],
feed_dict={query_idx: idx})
# Fail if the match_idx != idx, and the similarity of match_idx does
# is not tied with the argmax (which is 1.0 by construction).
if match_idx != idx:
self.assertClose(match_similarity, 1.0)
def testLibraryMatchingNoPredictions(self):
"""Test library_matching using hardcoded values with no predicted data."""
y_observed = np.array([[1, 2], [2, 1], [0, 0]], dtype=np.float32)
y_query = np.array([[2, 5], [-3, 1], [0, 0]], dtype=np.float32)
ids_observed = self.make_ids(3)
ids_query = self.make_ids(3)
expected_predictions = ['0', '2', '0']
masses_query = np.ones([3, 1], dtype=np.float32)
query_data = self._package_data(ids=ids_query,
spectrum=y_query,
masses=masses_query)
masses_observed = np.ones([3, 1], dtype=np.float32)
observed_data = self._package_data(ids=ids_observed,
spectrum=y_observed,
masses=masses_observed)
predicted_data = None
library_matching_data = library_matching.LibraryMatchingData(
query=query_data, observed=observed_data, predicted=predicted_data)
_, predicted_data, _, _ = library_matching.library_matching(
library_matching_data,
predictor_fn=None,
similarity_provider=similarity_lib.CosineSimilarityProvider(),
mass_tolerance=3.0)
with tf.Session() as sess:
sess.run(tf.local_variables_initializer())
predictions = sess.run(predicted_data[fmap_constants.INCHIKEY])
self.assertAllEqual(expected_predictions, predictions)
def perform_matching(self, ids_observed, ids_predicted, ids_query,
masses_observed, masses_predicted, masses_query,
y_observed, y_query, x_predicted, tf_transform,
mass_tolerance):
query_data = self._package_data(ids=ids_query,
spectrum=y_query,
masses=masses_query)
predicted_data = self._package_data(ids=ids_predicted,
spectrum=None,
masses=masses_predicted)
predicted_data[PREDICTOR_INPUT_KEY] = tf.constant(x_predicted)
observed_data = self._package_data(ids=ids_observed,
spectrum=y_observed,
masses=masses_observed)
library_matching_data = library_matching.LibraryMatchingData(
query=query_data, observed=observed_data, predicted=predicted_data)
predictor_fn = lambda d: tf_transform(d[PREDICTOR_INPUT_KEY])
similarity = similarity_lib.CosineSimilarityProvider()
true_data, predicted_data, _, _ = (library_matching.library_matching(
library_matching_data, predictor_fn, similarity, mass_tolerance,
10))
with tf.Session() as sess:
sess.run(tf.local_variables_initializer())
return sess.run([predicted_data, true_data])
def testLibraryMatchingNoObserved(self):
"""Test library_matching using hardcoded values with no observed data."""
tf_transform = lambda t: t + 2
x_predicted = np.array([[1, 1], [-3, -2]], dtype=np.float32)
y_query = np.array([[2, 5], [-3, 1], [0, 0]], dtype=np.float32)
ids_predicted = self.make_ids(2)
ids_query = self.make_ids(3)
expected_predictions = ['0', '1', '0']
masses_query = np.ones([3, 1], dtype=np.float32)
query_data = self._package_data(ids=ids_query,
spectrum=y_query,
masses=masses_query)
masses_predicted = np.ones([2, 1], dtype=np.float32)
predicted_data = self._package_data(ids=ids_predicted,
spectrum=None,
masses=masses_predicted)
predicted_data[PREDICTOR_INPUT_KEY] = tf.constant(x_predicted)
observed_data = None
library_matching_data = library_matching.LibraryMatchingData(
query=query_data, observed=observed_data, predicted=predicted_data)
predictor_fn = lambda d: tf_transform(d[PREDICTOR_INPUT_KEY])
_, predicted_data, _, _ = library_matching.library_matching(
library_matching_data,
predictor_fn=predictor_fn,
similarity_provider=similarity_lib.CosineSimilarityProvider(),
mass_tolerance=3.0)
with tf.Session() as sess:
sess.run(tf.local_variables_initializer())
predictions = sess.run(predicted_data[fmap_constants.INCHIKEY])
self.assertAllEqual(expected_predictions, predictions)