def test_init_raises(self):
six.assertRaisesRegex(self, TypeError,
'root must have at least one output',
tdc.Compiler.create, tdb.Record([]))
six.assertRaisesRegex(self, TypeError,
'root outputs must all be tensors',
tdc.Compiler.create, tdb.GetItem('foo'))
six.assertRaisesRegex(self, TypeError,
'root output may not contain sequences',
tdc.Compiler.create, tdb.Map(tdb.Scalar()))
def test_record_composition(self):
d = tdb.Record({'a': tdb.Scalar(), 'b': tdb.Scalar()})
fn1 = times_scalar_block(2.0)
fn2 = times_scalar_block(3.0)
fn3 = tdb.Function(tf.add)
c = tdb.Composition([d, fn1, fn2, fn3])
c.connect(c.input, d)
c.connect(d['a'], fn1)
c.connect(d['b'], fn2)
c.connect((fn1, fn2), fn3)
c.connect(fn3, c.output)
self.assertBuilds(17.0, c, {'a': 1.0, 'b': 5.0}, max_depth=2)
def test_forward_declarations(self):
# Define a simple expression data structure
nlit = lambda x: {'op': 'lit', 'val': x}
nadd = lambda x, y: {'op': 'add', 'left': x, 'right': y}
nexpr = nadd(nadd(nlit(3.0), nlit(5.0)), nlit(2.0))
# Define a recursive block using forward declarations
expr_fwd = tdb.ForwardDeclaration(tdt.PyObjectType(),
tdt.TensorType((), 'float32'))
lit_case = tdb.GetItem('val') >> tdb.Scalar()
add_case = (tdb.Record({'left': expr_fwd(), 'right': expr_fwd()})
>> tdb.Function(tf.add))
expr = tdb.OneOf(lambda x: x['op'], {'lit': lit_case, 'add': add_case})
expr_fwd.resolve_to(expr)
self.assertBuilds(10.0, expr, nexpr, max_depth=2)
def test_eval_empty_tuple(self): block = tdb.Record([]) self.assertBuildsConst((), block, ())
def test_length_empty_tuple(self): block = tdb.Record([]) >> tdb.Length(dtype='int32') self.assertBuildsConst(0, block, ())
def test_record_slice_key(self):
b = tdb.Record([
(0, tdb.Scalar()),
(slice(1, 3), (tdb.Scalar(), tdb.Scalar()) >> tdb.Concat())])
self.assertBuilds((1., [2., 3.]), b, [1, 2, 3])
def test_record_one_child(self):
self.assertBuildsConst((42,), tdb.Record({0: tdb.Scalar('int32')}), {0: 42})
def test_record(self):
d = tdb.Record(collections.OrderedDict([('b', tdb.Scalar()),
('a', tdb.Scalar())]))
c = d >> tdb.Function(tf.subtract)
self.assertBuilds(4.0, c, {'a': 1.0, 'b': 5.0})
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)
def test_eval_metrics_different_lengths(self):
b = tdb.Record((tdb.Map(tdb.Scalar('int32') >> tdm.Metric('x')),
tdb.Map(tdb.Scalar() >> tdm.Metric('y'))))
desired = ([None, None], [None]), {'x': [1, 2], 'y': [3.]}
self.assertBuilds(desired, b, ([1, 2], [3]))
