def _get_activations_for_batch(self,
list_of_seqs,
custom_tensor_to_retrieve=None):
"""Gets activations for each sequence in list_of_seqs.
[
[activation_1, activation_2, ...] # For list_of_seqs[0]
[activation_1, activation_2, ...] # For list_of_seqs[1]
...
]
In the case that the activations are the normalized probabilities that a
sequence belongs to a class, entry `i, j` of
`inferrer.get_activations(batch)` contains the probability that
sequence `i` is in family `j`.
Args:
list_of_seqs: list of strings, with characters that are amino acids.
custom_tensor_to_retrieve: string name for a tensor to retrieve, if unset
uses default for signature.
Returns:
np.array of floats containing the value from fetch_op.
"""
one_hots = [utils.residues_to_one_hot(seq) for seq in list_of_seqs]
padded_one_hots = utils.make_padded_np_array(one_hots)
tensor_name = custom_tensor_to_retrieve or self._fetch_tensor_name
with self._graph.as_default():
return self._sess.run(
tensor_name, {
self._signature.inputs[protein_dataset.SEQUENCE_KEY].name:
padded_one_hots,
self._signature.inputs[protein_dataset.SEQUENCE_LENGTH_KEY].name:
np.array([len(s) for s in list_of_seqs])
})
def testSparseToOneHot(self):
seq = 'AY'
expected_output = [[
1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0.
],
[
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0., 0., 0., 1.
]]
self.assertListEqual(expected_output,
utils.residues_to_one_hot(seq).tolist())