def test_save_model_to_tiledb_array_predictions(self, tmpdir, api, loss,
optimizer, metrics):
model = (api(num_hidden=1, num_classes=2, input_dim=3)
if api != ConfigSubclassModel else api(
hidden_units=[16, 16, 10]))
tiledb_uri = os.path.join(tmpdir, "model_array")
# Compiles the model if optimizer is present
if optimizer:
model.compile(loss=loss, optimizer=optimizer, metrics=[metrics])
input_shape = tuple(np.random.randint(20, size=2))
if not model.built:
model.build(input_shape)
tiledb_model_obj = TensorflowKerasTileDBModel(uri=tiledb_uri,
model=model)
tiledb_model_obj.save(include_optimizer=True if optimizer else False)
loaded_model = (tiledb_model_obj.load(
compile_model=False,
custom_objects={"ConfigSubclassModel": ConfigSubclassModel},
input_shape=input_shape,
) if api == ConfigSubclassModel else tiledb_model_obj.load(
compile_model=False))
data = np.random.rand(
100, input_shape[-1] if api == ConfigSubclassModel else 3)
# Assert model predictions are equal
np.testing.assert_array_equal(loaded_model.predict(data),
model.predict(data))
def test_sequential_model_save_load_without_input_shape(
self, tmpdir, api, loss, optimizer, metrics):
if optimizer is None or loss != keras.losses.MSE:
pytest.skip()
model = keras.models.Sequential()
model.add(keras.layers.Dense(2))
model.add(keras.layers.RepeatVector(3))
model.add(keras.layers.TimeDistributed(keras.layers.Dense(3)))
model.compile(
loss=loss,
optimizer=optimizer,
metrics=metrics,
weighted_metrics=metrics,
sample_weight_mode="temporal",
)
data_x = np.random.random((1, 3))
data_y = np.random.random((1, 3, 3))
model.train_on_batch(data_x, data_y)
tiledb_uri = os.path.join(tmpdir, "model_array")
tiledb_model_obj = TensorflowKerasTileDBModel(uri=tiledb_uri,
model=model)
tiledb_model_obj.save(include_optimizer=True)
loaded_model = tiledb_model_obj.load(compile_model=True)
# Assert model predictions are equal
np.testing.assert_array_equal(loaded_model.predict(data_x),
model.predict(data_x))
def test_save_load_for_rnn_layers(self, tmpdir, api, loss, optimizer,
metrics):
inputs = keras.Input([10, 10], name="train_input")
rnn_layers = [
keras.layers.LSTMCell(size,
recurrent_dropout=0,
name="rnn_cell%d" % i)
for i, size in enumerate([32, 32])
]
rnn_output = keras.layers.RNN(rnn_layers,
return_sequences=True,
name="rnn_layer")(inputs)
pred_feat = keras.layers.Dense(10,
name="prediction_features")(rnn_output)
pred = keras.layers.Softmax()(pred_feat)
model = keras.Model(inputs=[inputs], outputs=[pred, pred_feat])
tiledb_uri = os.path.join(tmpdir, "model_array")
tiledb_model_obj = TensorflowKerasTileDBModel(uri=tiledb_uri,
model=model)
tiledb_model_obj.save(include_optimizer=False)
loaded_model = tiledb_model_obj.load(compile_model=False)
data = np.random.rand(50, 10, 10)
# Assert model predictions are equal
np.testing.assert_array_equal(loaded_model.predict(data),
model.predict(data))
def test_save_load_with_dense_features(self, tmpdir, api, loss, optimizer,
metrics):
if optimizer is None:
pytest.skip()
cols = [
feature_column_lib.numeric_column("a"),
feature_column_lib.indicator_column(
feature_column_lib.categorical_column_with_vocabulary_list(
"b", ["one", "two"])),
]
input_layers = {
"a": keras.layers.Input(shape=(1, ), name="a"),
"b": keras.layers.Input(shape=(1, ), name="b", dtype="string"),
}
fc_layer = dense_features.DenseFeatures(cols)(input_layers)
output = keras.layers.Dense(10)(fc_layer)
model = keras.models.Model(input_layers, output)
model.compile(
loss=loss,
optimizer=optimizer,
metrics=[metrics],
)
tiledb_uri = os.path.join(tmpdir, "model_array")
tiledb_model_obj = TensorflowKerasTileDBModel(uri=tiledb_uri,
model=model)
tiledb_model_obj.save(include_optimizer=True)
loaded_model = tiledb_model_obj.load(compile_model=True)
model_opt_weights = batch_get_value(getattr(model.optimizer,
"weights"))
loaded_opt_weights = batch_get_value(
getattr(loaded_model.optimizer, "weights"))
# Assert optimizer weights are equal
for weight_model, weight_loaded_model in zip(model_opt_weights,
loaded_opt_weights):
np.testing.assert_array_equal(weight_model, weight_loaded_model)
inputs_a = np.arange(10).reshape(10, 1)
inputs_b = np.arange(10).reshape(10, 1).astype("str")
# Assert model predictions are equal
np.testing.assert_array_equal(
loaded_model.predict({
"a": inputs_a,
"b": inputs_b
}),
model.predict({
"a": inputs_a,
"b": inputs_b
}),
)
def test_functional_model_save_load_with_custom_loss_and_metric(
self, tmpdir, api, loss, optimizer, metrics):
if optimizer is None or loss != keras.losses.SparseCategoricalCrossentropy(
):
pytest.skip()
inputs = keras.Input(shape=(4, ))
x = keras.layers.Dense(8, activation="relu")(inputs)
outputs = keras.layers.Dense(3, activation="softmax")(x)
model = keras.Model(inputs=inputs, outputs=outputs)
custom_loss = keras.layers.Lambda(lambda x: keras.backend.sum(x * x))(
x)
model.add_loss(custom_loss)
model.add_metric(custom_loss, aggregation="mean", name="custom_loss")
model.compile(
loss=loss,
optimizer=optimizer,
metrics=[metrics],
)
data_x = np.random.normal(size=(32, 4))
data_y = np.random.randint(0, 3, size=32)
model.train_on_batch(data_x, data_y)
tiledb_uri = os.path.join(tmpdir, "model_array")
tiledb_model_obj = TensorflowKerasTileDBModel(uri=tiledb_uri,
model=model)
tiledb_model_obj.save(include_optimizer=True)
loaded_model = tiledb_model_obj.load(compile_model=True)
# Assert all evaluation results are the same.
assert all([
a == pytest.approx(b, 1e-9) for a, b in zip(
model.evaluate(data_x, data_y),
loaded_model.evaluate(data_x, data_y),
)
])
# Assert model predictions are equal
np.testing.assert_array_equal(loaded_model.predict(data_x),
model.predict(data_x))
def test_load_tiledb_error_with_wrong_uri():
tiledb_model_obj = TensorflowKerasTileDBModel(uri="dummy_uri")
with pytest.raises(tiledb.TileDBError):
tiledb_model_obj.load(compile_model=False)
def test_save_load_with_sequence_features(self, tmpdir, api, loss,
optimizer, metrics):
if optimizer is None:
pytest.skip()
cols = [
feature_column_lib.sequence_numeric_column("a"),
feature_column_lib.indicator_column(
feature_column_lib.
sequence_categorical_column_with_vocabulary_list(
"b", ["one", "two"])),
]
input_layers = {
"a":
keras.layers.Input(shape=(None, 1), sparse=True, name="a"),
"b":
keras.layers.Input(shape=(None, 1),
sparse=True,
name="b",
dtype="string"),
}
fc_layer, _ = ksfc.SequenceFeatures(cols)(input_layers)
x = keras.layers.GRU(32)(fc_layer)
output = keras.layers.Dense(10)(x)
model = keras.models.Model(input_layers, output)
model.compile(
loss=loss,
optimizer=optimizer,
metrics=[metrics],
)
tiledb_uri = os.path.join(tmpdir, "model_array")
tiledb_model_obj = TensorflowKerasTileDBModel(uri=tiledb_uri,
model=model)
tiledb_model_obj.save(include_optimizer=True)
loaded_model = tiledb_model_obj.load(compile_model=True)
model_opt_weights = batch_get_value(getattr(model.optimizer,
"weights"))
loaded_opt_weights = batch_get_value(
getattr(loaded_model.optimizer, "weights"))
# Assert optimizer weights are equal
for weight_model, weight_loaded_model in zip(model_opt_weights,
loaded_opt_weights):
np.testing.assert_array_equal(weight_model, weight_loaded_model)
batch_size = 10
timesteps = 1
values_a = np.arange(10, dtype=np.float32)
indices_a = np.zeros((10, 3), dtype=np.int64)
indices_a[:, 0] = np.arange(10)
inputs_a = sparse_tensor.SparseTensor(indices_a, values_a,
(batch_size, timesteps, 1))
values_b = np.zeros(10, dtype=np.str)
indices_b = np.zeros((10, 3), dtype=np.int64)
indices_b[:, 0] = np.arange(10)
inputs_b = sparse_tensor.SparseTensor(indices_b, values_b,
(batch_size, timesteps, 1))
# Assert model predictions are equal
np.testing.assert_array_equal(
loaded_model.predict({
"a": inputs_a,
"b": inputs_b
}, steps=1),
model.predict({
"a": inputs_a,
"b": inputs_b
}, steps=1),
)