def test_rnn_with_cells(self):
gru_cell1 = tdl.ScopedLayer(tf.contrib.rnn.GRUCell(num_units=16), 'gru1')
gru_cell2 = tdl.ScopedLayer(tf.contrib.rnn.GRUCell(num_units=16), 'gru2')
with tf.variable_scope('gru3') as vscope:
gru_cell3 = tdl.ScopedLayer(tf.contrib.rnn.GRUCell(num_units=16), vscope)
lstm_cell = tdl.ScopedLayer(
tf.contrib.rnn.BasicLSTMCell(num_units=16), 'lstm')
gru1 = (tdb.InputTransform(lambda s: [ord(c) for c in s]) >>
tdb.Map(tdb.Scalar('int32') >>
tdb.Function(tdl.Embedding(128, 8))) >>
tdb.RNN(gru_cell1))
gru2 = (tdb.InputTransform(lambda s: [ord(c) for c in s]) >>
tdb.Map(tdb.Scalar('int32') >>
tdb.Function(tdl.Embedding(128, 8))) >>
tdb.RNN(gru_cell2, initial_state=tf.ones(16)))
gru3 = (tdb.InputTransform(lambda s: [ord(c) for c in s]) >>
tdb.Map(tdb.Scalar('int32') >>
tdb.Function(tdl.Embedding(128, 8))) >>
tdb.RNN(gru_cell3, initial_state=tdb.FromTensor(tf.ones(16))))
lstm = (tdb.InputTransform(lambda s: [ord(c) for c in s]) >>
tdb.Map(tdb.Scalar('int32') >>
tdb.Function(tdl.Embedding(128, 8))) >>
tdb.RNN(lstm_cell))
with self.test_session():
gru1.eval('abcde')
gru2.eval('ABCDE')
gru3.eval('vghj')
lstm.eval('123abc')
def test_forward_declaration_orphaned_nested(self):
fwd1 = tdb.ForwardDeclaration(tdt.VoidType(), tdt.TensorType([]))
fwd2 = tdb.ForwardDeclaration(tdt.SequenceType(tdt.TensorType([])),
tdt.TensorType([]))
b = tdb.Map(tdb.Scalar()) >> fwd2() >> tdb.Function(tf.negative)
fwd2.resolve_to(tdb.Fold(tdb.Function(tf.add), fwd1()))
fwd1.resolve_to(tdb.FromTensor(tf.ones([])))
self.assertBuilds(-8., b, [3, 4], max_depth=3)
def test_from_tensor_variable(self): v = tf.Variable(np.ones(2)) block = tdb.FromTensor(v) self.assertBuilds([1., 1.], block, None, max_depth=0)
def test_from_tensor_const_ndarray(self): block = tdb.FromTensor(np.zeros([2, 2], 'int32')) self.assertBuildsConst([[0, 0], [0, 0]], block, None)
def test_from_tensor_const(self): block = tdb.FromTensor(tf.constant([1, 2, 3], 'int32')) self.assertBuildsConst([1, 2, 3], block, None)
def test_from_tensor(self):
t = tf.placeholder('int32', [2, 1])
block = tdb.FromTensor(t)
fd = {t: [[1], [2]]}
self.assertBuilds([[1], [2]], block, None, max_depth=0, feed_dict=fd)
def test_forward_declaration_orphaned(self): fwd = tdb.ForwardDeclaration(tdt.VoidType(), tdt.TensorType([])) b = tdb.AllOf(fwd(), fwd()) >> tdb.Sum() fwd.resolve_to(tdb.FromTensor(tf.ones([]))) self.assertBuilds(2., b, None)
def test_repr(self):
goldens = {
tdb.Tensor([]): '<td.Tensor dtype=\'float32\' shape=()>',
tdb.Tensor([1, 2], 'int32', name='foo'):
'<td.Tensor \'foo\' dtype=\'int32\' shape=(1, 2)>',
tdb.Scalar('int64'): '<td.Scalar dtype=\'int64\'>',
tdb.Vector(42): '<td.Vector dtype=\'float32\' size=42>',
tdb.FromTensor(tf.zeros(3)): '<td.FromTensor \'zeros:0\'>',
tdb.Function(tf.negative,
name='foo'): '<td.Function \'foo\' tf_fn=\'negative\'>',
tdb.Identity(): '<td.Identity>',
tdb.Identity('foo'): '<td.Identity \'foo\'>',
tdb.InputTransform(ord): '<td.InputTransform py_fn=\'ord\'>',
tdb.SerializedMessageToTree('foo'):
'<td.SerializedMessageToTree \'foo\' '
'py_fn=\'serialized_message_to_tree\'>',
tdb.GetItem(3, 'mu'): '<td.GetItem \'mu\' key=3>',
tdb.Length(): '<td.Length dtype=\'float32\'>',
tdb.Slice(stop=2): '<td.Slice key=slice(None, 2, None)>',
tdb.Slice(stop=2, name='x'):
'<td.Slice \'x\' key=slice(None, 2, None)>',
tdb.ForwardDeclaration(name='foo')():
'<td.ForwardDeclaration() \'foo\'>',
tdb.Composition(name='x').input: '<td.Composition.input \'x\'>',
tdb.Composition(name='x').output: '<td.Composition.output \'x\'>',
tdb.Composition(name='x'): '<td.Composition \'x\'>',
tdb.Pipe(): '<td.Pipe>',
tdb.Pipe(tdb.Scalar(), tdb.Identity()): '<td.Pipe>',
tdb.Record({}, name='x'): '<td.Record \'x\' ordered=False>',
tdb.Record((), name='x'): '<td.Record \'x\' ordered=True>',
tdb.AllOf(): '<td.AllOf>',
tdb.AllOf(tdb.Identity()): '<td.AllOf>',
tdb.AllOf(tdb.Identity(), tdb.Identity()): '<td.AllOf>',
tdb.AllOf(name='x'): '<td.AllOf \'x\'>',
tdb.AllOf(tdb.Identity(), name='x'): '<td.AllOf \'x\'>',
tdb.AllOf(tdb.Identity(), tdb.Identity(), name='x'): '<td.AllOf \'x\'>',
tdb.Map(tdb.Scalar(), name='x'):
'<td.Map \'x\' element_block=<td.Scalar dtype=\'float32\'>>',
tdb.Fold(tdb.Function(tf.add), tf.ones([]), name='x'):
'<td.Fold \'x\' combine_block=<td.Function tf_fn=\'add\'> '
'start_block=<td.FromTensor \'ones:0\'>>',
tdb.RNN(tdl.ScopedLayer(tf.contrib.rnn.GRUCell(num_units=8))):
'<td.RNN>',
tdb.RNN(tdl.ScopedLayer(tf.contrib.rnn.GRUCell(num_units=8)), name='x'):
'<td.RNN \'x\'>',
tdb.RNN(tdl.ScopedLayer(tf.contrib.rnn.GRUCell(num_units=8)),
initial_state=tf.ones(8)):
'<td.RNN>',
tdb.RNN(tdl.ScopedLayer(tf.contrib.rnn.GRUCell(num_units=8)),
initial_state=tf.ones(8), name='x'):
'<td.RNN \'x\'>',
tdb.Reduce(tdb.Function(tf.add), name='x'):
'<td.Reduce \'x\' combine_block=<td.Function tf_fn=\'add\'>>',
tdb.Sum(name='foo'):
'<td.Sum \'foo\' combine_block=<td.Function tf_fn=\'add\'>>',
tdb.Min(name='foo'):
'<td.Min \'foo\' combine_block=<td.Function tf_fn=\'minimum\'>>',
tdb.Max(name='foo'):
'<td.Max \'foo\' combine_block=<td.Function tf_fn=\'maximum\'>>',
tdb.Mean(name='foo'): '<td.Mean \'foo\'>',
tdb.OneOf(ord, (tdb.Scalar(), tdb.Scalar()), name='x'):
'<td.OneOf \'x\'>',
tdb.Optional(tdb.Scalar(), name='foo'):
'<td.Optional \'foo\' some_case_block=<td.Scalar dtype=\'float32\'>>',
tdb.Concat(1, True, 'x'):
'<td.Concat \'x\' concat_dim=1 flatten=True>',
tdb.Broadcast(name='x'): '<td.Broadcast \'x\'>',
tdb.Zip(name='x'): '<td.Zip \'x\'>',
tdb.NGrams(n=42, name='x'): '<td.NGrams \'x\' n=42>',
tdb.OneHot(2, 3, name='x'):
'<td.OneHot \'x\' dtype=\'float32\' start=2 stop=3>',
tdb.OneHot(3): '<td.OneHot dtype=\'float32\' start=0 stop=3>',
tdb.OneHotFromList(['a', 'b']): '<td.OneHotFromList>',
tdb.OneHotFromList(['a', 'b'], name='foo'):
'<td.OneHotFromList \'foo\'>',
tdb.Nth(name='x'): '<td.Nth \'x\'>',
tdb.Zeros([], 'x'): '<td.Zeros \'x\'>',
tdb.Void(): '<td.Void>',
tdb.Void('foo'): '<td.Void \'foo\'>',
tdm.Metric('foo'): '<td.Metric \'foo\'>'}
for block, expected_repr in sorted(six.iteritems(goldens),
key=lambda kv: kv[1]):
self.assertEqual(repr(block), expected_repr)