def test_sequence_coder(self):
coder = serialization.SequenceCoder(
specs={
'seq': tf.float32,
'name': tf.string,
'multi': tf.float32,
'ints': tf.int64,
},
sequence_keys=['seq', 'multi', 'ints'])
encoded = coder.encode(self.example)
decoded = coder.decode(encoded)
reencoded = coder.encode(decoded)
redecoded = coder.decode(reencoded)
self.assertAllClose(redecoded['seq'][:, 0],
self.example['seq'],
atol=1e-7)
self.assertAllClose(redecoded['seq'], decoded['seq'], atol=1e-7)
self.assertAllClose(redecoded['multi'],
self.example['multi'],
atol=1e-7)
self.assertAllClose(redecoded['ints'][:, 0],
self.example['ints'],
atol=1e-7)
self.assertEqual(redecoded['name'], self.example['name'])
def get_sequence_coder(specs):
"""Returns a SequenceCoder with the proper specs."""
def not_sequence(v):
return v == tf.string or isinstance(v, tf.io.FixedLenFeature)
sequence_keys = [k for k, v in specs.items() if not not_sequence(v)]
return serialization.SequenceCoder(specs=specs,
sequence_keys=sequence_keys)