def setUpClass(cls):
client = grakn.client.GraknClient(uri="localhost:48555")
cls.session = client.session(keyspace="test_schema")
entity_query = "match $x isa company, has name 'Google'; get;"
cls._tx = cls.session.transaction().write()
neighbour_sample_sizes = (4, 3)
sampling_method = ordered.ordered_sample
samplers = []
for sample_size in neighbour_sample_sizes:
samplers.append(
samp.Sampler(sample_size,
sampling_method,
limit=sample_size * 2))
grakn_thing = next(cls._tx.query(entity_query)).get('x')
thing = neighbour.build_thing(grakn_thing)
context_builder = builder.ContextBuilder(samplers)
cls.context = context_builder.build(cls._tx, thing)
def test_neighbour_finder_called_with_root_and_neighbour_ids(self):
tx_mock = mock.Mock(grakn.client.Transaction)
sampler = mock.Mock(samp.Sampler)
sampler.return_value = mocks.gen([
mock.MagicMock(neighbour.Connection,
role_label="employmee",
role_direction=1,
neighbour_thing=mock.MagicMock(neighbour.Thing,
id="1")),
mock.MagicMock(neighbour.Connection,
role_label="@has-name-owner",
role_direction=1,
neighbour_thing=mock.MagicMock(neighbour.Thing,
id="3")),
])
sampler2 = mock.Mock(samp.Sampler)
sampler2.return_value = []
starting_thing = mock.MagicMock(neighbour.Thing, id="0")
mock_neighbour_finder = mock.MagicMock(neighbour.NeighbourFinder)
context_builder = builder.ContextBuilder(
[sampler, sampler2], neighbour_finder=mock_neighbour_finder)
# The call to assess
context_builder.build(tx_mock, starting_thing)
print(mock_neighbour_finder.find.mock_calls)
mock_neighbour_finder.find.assert_has_calls([
mock.call("0", tx_mock),
mock.call("1", tx_mock),
mock.call("3", tx_mock)
])
def test_build_context_for_1_hop(self):
starting_thing = neighbour.Thing("0", "person", "entity")
samplers = [samp.Sampler(2, ordered.ordered_sample, limit=2)]
context_builder = builder.ContextBuilder(
samplers, neighbour_finder=mocks.MockNeighbourFinder())
thing_context = context_builder.build(
mock.Mock(grakn.client.Transaction), starting_thing)
expected_context = {
0: [
builder.Node((0, ),
neighbour.Thing("1",
"name",
"attribute",
data_type='string',
value='Sundar Pichai'), "has",
neighbour.NEIGHBOUR_PLAYS),
builder.Node((1, ),
neighbour.Thing("2", "employment", "relation"),
"employee", neighbour.TARGET_PLAYS),
],
1: [builder.Node((), neighbour.Thing("0", "person", "entity"))],
}
self.assertEqual(expected_context, thing_context)
def setUp(self):
entity_query = "match $x isa company, has name 'Google'; get;"
uri = "localhost:48555"
keyspace = "test_schema"
client = grakn.Grakn(uri=uri)
session = client.session(keyspace=keyspace)
self._tx = session.transaction(grakn.TxType.WRITE)
neighbour_sample_sizes = (4, 3)
sampling_method = ordered.ordered_sample
samplers = []
for sample_size in neighbour_sample_sizes:
samplers.append(samp.Sampler(sample_size, sampling_method, limit=sample_size * 2))
grakn_things = [answermap.get('x') for answermap in list(self._tx.query(entity_query))]
things = [neighbour.build_thing(grakn_thing) for grakn_thing in grakn_things]
context_builder = builder.ContextBuilder(samplers)
self._neighbourhood_depths = [context_builder.build(self._tx, thing) for thing in things]
self._neighbour_roles = builder.convert_thing_contexts_to_neighbours(self._neighbourhood_depths)
self._flattened = flatten_tree(self._neighbour_roles)
def _neighbourhood_traverser_factory(neighbour_sample_sizes):
sampling_method = ordered.ordered_sample
samplers = []
for sample_size in neighbour_sample_sizes:
samplers.append(samp.Sampler(sample_size, sampling_method, limit=sample_size * 2))
context_builder = builder.ContextBuilder(samplers)
return context_builder
def __init__(
self,
neighbour_sample_sizes,
features_size,
example_concepts_features_size,
aggregated_size,
embedding_size,
schema_encoding_transaction,
batch_size,
embedding_normalisation=tf.nn.l2_normalize,
neighbour_sampling_method=ordered.ordered_sample,
neighbour_sampling_limit_factor=1,
formatters={'neighbour_value_date': preprocess.datetime_to_unixtime},
features_to_exclude=()):
self._embedding_normalisation = embedding_normalisation
self.embedding_size = embedding_size
self.aggregated_size = aggregated_size
self.neighbour_sample_sizes = neighbour_sample_sizes
self.feature_sizes = [features_size] * len(self.neighbour_sample_sizes)
self.feature_sizes[-1] = example_concepts_features_size
print(f'feature sizes: {self.feature_sizes}')
self._schema_encoding_transaction = schema_encoding_transaction
self._encode = encode.Encoder(self._schema_encoding_transaction)
self.batch_size = batch_size
self._formatters = formatters
self._features_to_exclude = features_to_exclude
traversal_samplers = []
for sample_size in neighbour_sample_sizes:
traversal_samplers.append(
sample.Sampler(sample_size,
neighbour_sampling_method,
limit=int(sample_size *
neighbour_sampling_limit_factor)))
self._array_builder = array.ArrayConverter(neighbour_sample_sizes)
self._context_builder = builder.ContextBuilder(traversal_samplers)
self._embed = embed.Embedder(
self.feature_sizes,
self.aggregated_size,
self.embedding_size,
self.neighbour_sample_sizes,
normalisation=self._embedding_normalisation)
features_to_exclude = {
feat_name: None
for feat_name in self._features_to_exclude
}
self.neighbourhood_dataset, self.array_placeholders = preprocess.build_dataset(
self.neighbour_sample_sizes, **features_to_exclude)
def test_build_context_for_0_hop(self):
starting_thing = neighbour.Thing("0", "person", "entity")
samplers = []
context_builder = builder.ContextBuilder(samplers, neighbour_finder=mocks.MockNeighbourFinder())
thing_context = context_builder.build(mock.Mock(grakn.client.Transaction), starting_thing)
expected_context = {
0: [builder.Node((), neighbour.Thing("0", "person", "entity"))],
}
self.assertEqual(expected_context, thing_context)
def test_input_output(self):
neighbour_sample_sizes = (2, 3)
samplers = [lambda x: x for sample_size in neighbour_sample_sizes]
starting_thing = neighbour.Thing("0", "person", "entity")
context_builder = builder.ContextBuilder(samplers, neighbour_finder=mocks.MockNeighbourFinder())
thing_context = context_builder.build(self._tx, starting_thing)
self.assertEqual(thing_context, mocks.mock_traversal_output())
def test_build_context_batch(self):
batch = [neighbour.Thing("0", "person", "entity"), neighbour.Thing("0", "person", "entity")]
samplers = []
context_builder = builder.ContextBuilder(samplers, neighbour_finder=mocks.MockNeighbourFinder())
thing_context = context_builder.build_batch(mock.Mock(grakn.client.Session), batch)
expected_context_batch = [{
0: [builder.Node((), neighbour.Thing("0", "person", "entity"))],
},
{
0: [builder.Node((), neighbour.Thing("0", "person", "entity"))],
}
]
self.assertEqual(expected_context_batch, thing_context)
def test_neighbour_finder_called_with_root_node_id(self):
tx_mock = mock.Mock(grakn.client.Transaction)
sampler = mock.Mock(samp.Sampler)
sampler.return_value = []
starting_thing = mock.MagicMock(neighbour.Thing, id="0")
mock_neighbour_finder = mock.MagicMock(neighbour.NeighbourFinder)
context_builder = builder.ContextBuilder(
[sampler], neighbour_finder=mock_neighbour_finder)
# The call to assess
context_builder.build(tx_mock, starting_thing)
mock_neighbour_finder.find.assert_called_once_with("0", tx_mock)
def setUp(self):
self._tx = self.session.transaction(grakn.TxType.WRITE)
neighbour_sample_sizes = (2, 3)
samplers = [lambda x: x for sample_size in neighbour_sample_sizes]
starting_thing = neighbour.Thing("0", "person", "entity")
things = [starting_thing]
context_builder = builder.ContextBuilder(samplers, neighbour_finder=mocks.MockNeighbourFinder())
self._neighbourhood_depths = [context_builder.build(self._tx, thing) for thing in things]
self._neighbour_roles = builder.convert_thing_contexts_to_neighbours(self._neighbourhood_depths)
self._flattened = flatten_tree(self._neighbour_roles)
def test_input_output_integration(self):
"""
Runs using real samplers
:return:
"""
sampling_method = ordered.ordered_sample
samplers = [samp.Sampler(2, sampling_method, limit=2), samp.Sampler(3, sampling_method, limit=1)]
starting_thing = neighbour.Thing("0", "person", "entity")
context_builder = builder.ContextBuilder(samplers, neighbour_finder=mocks.MockNeighbourFinder())
thing_context = context_builder.build(self._tx, starting_thing)
self.assertEqual(thing_context, mocks.mock_traversal_output())
def test_build_context_for_3_hop(self):
starting_thing = neighbour.Thing("0", "person", "entity")
samplers = [samp.Sampler(2, ordered.ordered_sample, limit=2), samp.Sampler(2, ordered.ordered_sample, limit=2),
samp.Sampler(3, ordered.ordered_sample, limit=3)]
context_builder = builder.ContextBuilder(samplers, neighbour_finder=mocks.MockNeighbourFinder())
thing_context = context_builder.build(mock.Mock(grakn.client.Transaction), starting_thing)
expected_context = {
3: [builder.Node((), neighbour.Thing("0", "person", "entity"))],
2: [builder.Node((0,), neighbour.Thing("1", "name", "attribute", data_type='string', value='Sundar Pichai'),
"has", neighbour.NEIGHBOUR_PLAYS),
builder.Node((1,), neighbour.Thing("2", "employment", "relation"), "employee", neighbour.TARGET_PLAYS),
],
1: [builder.Node((0, 0), neighbour.Thing("0", "person", "entity"), "has", neighbour.TARGET_PLAYS),
# Note that (0, 1) is reversed compared to the natural expectation
builder.Node((0, 1), neighbour.Thing("3", "company", "entity"), "employer", neighbour.NEIGHBOUR_PLAYS),
builder.Node((1, 1), neighbour.Thing("0", "person", "entity"), "employee", neighbour.NEIGHBOUR_PLAYS),
],
0: [builder.Node((0, 0, 0), neighbour.Thing("1", "name", "attribute", data_type='string', value='Sundar Pichai'),
"has", neighbour.NEIGHBOUR_PLAYS),
builder.Node((1, 0, 0), neighbour.Thing("2", "employment", "relation"), "employee", neighbour.TARGET_PLAYS),
builder.Node((0, 0, 1), neighbour.Thing("4", "name", "attribute", data_type='string', value='Google'),
"has", neighbour.NEIGHBOUR_PLAYS),
builder.Node((1, 0, 1), neighbour.Thing("4", "name", "attribute", data_type='string', value='Google'),
"has", neighbour.NEIGHBOUR_PLAYS),
builder.Node((2, 0, 1), neighbour.Thing("4", "name", "attribute", data_type='string', value='Google'),
"has", neighbour.NEIGHBOUR_PLAYS),
builder.Node((0, 1, 1), neighbour.Thing("1", "name", "attribute", data_type='string', value='Sundar Pichai'),
"has", neighbour.NEIGHBOUR_PLAYS),
builder.Node((1, 1, 1), neighbour.Thing("2", "employment", "relation"), "employee", neighbour.TARGET_PLAYS),
]
}
self.assertEqual(expected_context, thing_context)
