def test_set_slot_to_optimizer(self):
embed_name = "test_emb"
indices = np.ndarray([2], dtype=np.int32)
embed_values = np.ndarray([2, 2], dtype=np.float32)
slot_values = {
"m": np.ndarray([2, 2], dtype=np.float32),
"v": np.ndarray([2, 2], dtype=np.float32),
}
params = Parameters()
params.embedding_params[embed_name] = EmbeddingTable(embed_name, 8)
for slot in ["m", "v"]:
slot_table_name = get_slot_table_name(embed_name, slot)
params.embedding_params[slot_table_name] = EmbeddingTable(
slot_table_name, 2, "0.0", True)
opt = Adam()
opt_wrapper = OptimizerWrapper(opt, None, params.get_embedding_param)
opt_wrapper._init_thread_local()
opt_wrapper._tls._unique_ids_all_layers[embed_name] = indices
opt_wrapper._create_embedding_variable(embed_name, embed_values)
opt_wrapper._get_slot_and_set_to_optimizer(embed_name)
self.assertEqual(len(opt._slots), 1)
opt_slots = list(opt._slots.values())[0]
self.assertEqual(sorted(opt_slots.keys()), ["m", "v"])
for name in ["m", "v"]:
self.assertTrue(
np.allclose(opt_slots[name].numpy(), slot_values[name]))
def test_delete_variables(self):
params = Parameters()
embed_layers = ["test_1", "test_2"]
slot_names = ["m", "v"]
dim = 8
for layer in embed_layers:
params.embedding_params[layer] = EmbeddingTable(layer, dim)
for slot in slot_names:
slot_key = get_slot_table_name(layer, slot)
params.embedding_params[slot_key] = EmbeddingTable(
slot_key, dim, "0.0", True)
opt = Adam()
opt_wrapper = OptimizerWrapper(opt, None, params.get_embedding_param,
params.set_embedding_param)
opt_wrapper._init_thread_local()
for name in embed_layers:
opt_wrapper._tls._unique_ids_all_layers[name] = np.ndarray(
[2], np.int32)
opt_wrapper._create_embedding_variable(
name, np.ndarray([2, dim], np.float32))
opt_wrapper._get_slot_and_set_to_optimizer(name)
self.assertTrue(len(opt._weights) == 4)
self.assertTrue(len(opt._slots) == 2)
for slot_dict in opt._slots.values():
self.assertTrue(len(slot_dict) == 2)
opt_wrapper._delete_slots_and_weights_in_optimizer()
self.assertTrue(len(opt._weights) == 0)
self.assertTrue(len(opt._slots) == 0)
def test_update_embedding_param(self):
params = Parameters()
for name in ["test_1", "test_2"]:
params.embedding_params[name] = EmbeddingTable(name, 8)
slot_key = get_slot_table_name(name, "momentum")
params.embedding_params[slot_key] = EmbeddingTable(
slot_key, 8, "0.0", True)
indices = {
"test_1": np.array([1, 5]),
"test_2": np.array([10]),
}
embed_vars = {
"test_1": tf.Variable(np.random.rand(2, 8).astype(np.float32)),
"test_2": tf.Variable(np.random.rand(1, 8).astype(np.float32)),
}
slot_vars = {
"test_1": {
"momentum":
tf.Variable(np.random.rand(2, 8).astype(np.float32))
},
"test_2": {
"momentum":
tf.Variable(np.random.rand(1, 8).astype(np.float32))
},
}
opt = SGD(momentum=0.1)
opt_wrapper = OptimizerWrapper(opt, None, None,
params.set_embedding_param)
opt_wrapper._tls._unique_ids_all_layers = indices
opt_wrapper._tls._embed_variables = embed_vars
opt_wrapper._tls._slot_variables = slot_vars
opt_wrapper._update_embedding_param()
for name in ["test_1", "test_2"]:
self.assertTrue(
np.allclose(
embed_vars[name].numpy(),
params.get_embedding_param(name, indices[name]),
))
slot = "momentum"
slot_table_name = get_slot_table_name(name, slot)
self.assertTrue(
np.allclose(
slot_vars[name][slot].numpy(),
params.get_embedding_param(slot_table_name, indices[name]),
))
def _test_correctness(self, optimizer_class, X, Y, seed, **opt_kwargs):
"""Test the correctness of specific TensorFlow optimizer."""
_model_file = get_module_file_path(
os.path.dirname(os.path.realpath(__file__)),
"embedding_test_module.KerasEmbeddingModel",
)
model_module = load_module(_model_file).__dict__
# train model with TensorFlow optimizer
dim = 4
weights = self._random_init_model_weight([(4, dim), (4, dim), (72, 1),
(1, )], seed)
loss_fn = model_module["loss"]
model1 = model_module["KerasEmbeddingModel"](4, dim, weights)
opt1 = optimizer_class(**opt_kwargs)
_train(model1, opt1, X, Y, loss_fn, random_seed=seed)
model2 = model_module["EdlEmbeddingModel"](dim, weights[2:])
opt2 = optimizer_class(**opt_kwargs)
embedding_weight_names = [
layer.embedding_weight_name
for layer in find_layer(model2, Embedding)
]
# create Parameters object and initialize embedding vectors
params = Parameters()
for weight_name, embed_value in zip(embedding_weight_names,
weights[:2]):
embed_table = EmbeddingTable(weight_name, dim)
embed_table.set(range(len(embed_value)), embed_value)
params.embedding_params[weight_name] = embed_table
_train_edl_embedding_with_optimizer_wrapper(model2,
opt2,
X,
Y,
loss_fn,
params,
random_seed=seed)
# compare trained parameters
wrong_msg = (
"The updated parameters of Optimizer Wrapper and TensorFlow "
"optimizer %s differ." % opt1.get_config()["name"])
for layer1, layer2 in zip(model1.layers, model2.layers):
if "embedding" in layer2.name:
w1 = layer1.weights[0].numpy()
w2 = params.get_embedding_param(layer2.embedding_weight_name,
range(4))
self.assertTrue(np.isclose(w1, w2).all(), msg=wrong_msg)
else:
for w1, w2 in zip(layer1.weights, layer2.weights):
self.assertTrue(np.isclose(w1.numpy(), w2.numpy()).all(),
msg=wrong_msg)
def test_worker_pull_embedding(self):
model_def = "mnist_functional_api.mnist_functional_api.custom_model"
self._create_pserver(model_def, 2)
arguments = [
"--worker_id",
0,
"--job_type",
elasticdl_pb2.TRAINING,
"--minibatch_size",
self._batch_size,
"--model_zoo",
self._model_zoo_path,
"--model_def",
model_def,
"--distribution_strategy",
DistributionStrategy.PARAMETER_SERVER,
]
args = parse_worker_args(arguments)
worker = Worker(args, ps_channels=self._channels)
# Test lookup embedding vectors that do not exist
layers = ["test-2", "test-2-slot"]
ids = [3, 5, 1, 6, 10, 2, 1, 2, 4, 7, 9]
embedding_table_args = [
(layers[0], 8, "uniform", False),
(layers[1], 8, 3.3, True),
]
# initialize embedding table object
for pserver in self._pservers:
for layer, table_args in zip(layers, embedding_table_args):
pserver.parameters.embedding_params[layer] = EmbeddingTable(
*table_args
)
result_dict = {}
for layer in layers:
embedding = worker.pull_embedding_vectors(layer, ids)
result_dict[layer] = embedding
for layer in layers:
expected_result = []
for embedding_id in ids:
ps_id = int_to_id(embedding_id, len(self._pservers))
table = self._pservers[ps_id].parameters.embedding_params[
layer
]
expected_result.append(table.get([embedding_id]))
expected_result = np.concatenate(expected_result)
self.assertTrue(np.allclose(expected_result, result_dict[layer]))
def create_slot_params(self, slot_names, init_values):
embed_layer_names = list(self.embedding_params.keys())
for layer_name in embed_layer_names:
for slot_name in slot_names:
key = get_slot_table_name(layer_name, slot_name)
if key in self.embedding_params:
raise ValueError(
"An embedding layer has unexpected name %s" % key)
self.embedding_params[key] = EmbeddingTable(
key,
self.embedding_params[layer_name].dim,
init_values[slot_name],
True,
)
def test_create_embedding_table_for_slots(self):
slot_name = "momentum"
init_value = 3.5
table = EmbeddingTable(
get_slot_table_name(self.name, slot_name),
dim=self.dim,
initializer=init_value,
is_slot=True,
)
self.assertIsNotNone(table)
self.assertEqual(table.name, get_slot_table_name(self.name, slot_name))
self.assertEqual(table.dim, self.dim)
# test initialize
embedding = table.get([2])
self.assertTrue((embedding - init_value < 0.0001).all())
def test_save_parameters_to_checkpoint_file(self):
with tempfile.TemporaryDirectory() as tempdir:
checkpoint_saver = CheckpointSaver(
checkpoint_dir=os.path.join(tempdir, "ckpt/"),
checkpoint_steps=5,
keep_checkpoint_max=3,
include_evaluation=False,
)
pserver_servicer = PserverServicer(
parameters=Parameters(),
grads_to_wait=0,
optimizer="optimizer",
checkpoint_saver=checkpoint_saver,
ps_id=0,
num_ps_pods=1,
)
model_params = {
"v0": tf.Variable([[1, 1, 1], [1, 1, 1]]),
"v1": tf.Variable([[2, 2, 2], [2, 2, 2]]),
}
server_params = pserver_servicer._parameters
for var_name, var_value in model_params.items():
server_params.non_embedding_params[var_name] = var_value
embedding_table = EmbeddingTable(
name="embedding_0", dim=3, initializer="random_uniform"
)
server_params.embedding_params["embedding_0"] = embedding_table
server_params.set_embedding_param(
name="embedding_0",
indices=np.array([0, 1]),
values=np.array([[1, 1, 1], [2, 2, 2]]),
)
for i in range(100):
pserver_servicer._parameters.version += 1
pserver_servicer._save_params_to_checkpoint_if_needed()
self.assertEqual(len(os.listdir(checkpoint_saver._directory)), 3)
self.assertEqual(
sorted(os.listdir(checkpoint_saver._directory)),
["version-100", "version-90", "version-95"],
)
self.assertEqual(
os.listdir(checkpoint_saver._directory + "/version-100"),
["variables-0-of-1.ckpt"],
)
def _mock_model_parameters(self, model):
params = Parameters()
for weight in model.trainable_variables:
if "embedding" in weight.name:
embedding_table = EmbeddingTable(
name=weight.name,
dim=weight.shape[1],
initializer="RandomUniform",
)
embedding_table.set(np.arange(weight.shape[0]),
np.ones(weight.shape))
params.embedding_params[weight.name] = embedding_table
else:
params.non_embedding_params[weight.name] = tf.ones(
weight.shape)
params.version = 100
return params
def test_restore_parameters_from_checkpoint(self):
checkpoint_dir = "elasticdl/python/tests/testdata/ps_ckpt"
checkpoint_saver = CheckpointSaver(checkpoint_dir, 0, 0, False)
params = Parameters()
table = EmbeddingTable("embedding", 2, "random_uniform")
table.set([0, 1, 2, 3], np.ones((4, 2), dtype=np.float32))
params.embedding_params["embedding"] = table
params.non_embedding_params["dense/kernel:0"] = tf.Variable(
[[1.0], [1.0]]
)
params.non_embedding_params["dense/bias:0"] = tf.Variable([1.0])
params.version = 100
model_pb = params.to_model_pb()
checkpoint_saver.save(100, model_pb, False)
checkpoint_dir_for_init = checkpoint_dir + "/version-100"
args = PserverArgs(
ps_id=0,
num_ps_pods=2,
model_zoo=_test_model_zoo_path,
model_def="test_module.custom_model",
checkpoint_dir_for_init=checkpoint_dir_for_init,
)
pserver_0 = ParameterServer(args)
embedding_table = pserver_0.parameters.embedding_params["embedding"]
self.assertEqual(
list(embedding_table.embedding_vectors.keys()), [0, 2]
)
self.assertEqual(
list(pserver_0.parameters.non_embedding_params.keys()),
["dense/kernel:0"],
)
self.assertTrue(
np.array_equal(
pserver_0.parameters.non_embedding_params[
"dense/kernel:0"
].numpy(),
np.array([[1], [1]], dtype=int),
)
)
self.assertEqual(pserver_0.parameters.version, 100)
args = PserverArgs(
ps_id=1,
num_ps_pods=2,
model_zoo=_test_model_zoo_path,
model_def="test_module.custom_model",
checkpoint_dir_for_init=checkpoint_dir_for_init,
)
pserver_1 = ParameterServer(args)
embedding_table = pserver_1.parameters.embedding_params["embedding"]
self.assertEqual(
list(embedding_table.embedding_vectors.keys()), [1, 3]
)
self.assertEqual(
list(pserver_1.parameters.non_embedding_params.keys()),
["dense/bias:0"],
)
self.assertTrue(
np.array_equal(
pserver_1.parameters.non_embedding_params[
"dense/bias:0"
].numpy(),
np.array([1], dtype=int),
)
)
self.assertEqual(pserver_1.parameters.version, 100)
def _test_async_correctness(
self,
grads_and_vars_batches,
embed_values,
expected_non_embed_values,
expected_embed_values=None,
):
"""Checks the correctness of async OptimizerWrapper. This function
creates many threads and these threads call
`OptimizerWrapper.apply_gradients` simultaneously.
Args:
grads_and_vars_batches: A python list of `grads_and_vars`. Every
thread takes a `grads_and_vars` and calls `apply_gradients`.
embed_values: A python dictionary of
`(layer_name, embedding table)`.
expected_non_embed_values: A python list of expected non-embdding
values after applying gradients.
expected_embed_values: A python dictionary of expected embedding
values after applying gradients. None means no need to check
embedding values.
"""
thread_num = len(grads_and_vars_batches)
input_dims = {}
embed_var_n = len(embed_values)
params = Parameters()
for layer, values in embed_values.items():
embed_dim = values.shape[1]
input_dims[layer] = values.shape[0]
embed_table = EmbeddingTable(layer, embed_dim)
embed_table.set(range(input_dims[layer]), values)
params.embedding_params[layer] = embed_table
opt = SGD(0.1)
opt_wrapper = OptimizerWrapper(
opt,
True,
lookup_embedding_func=params.get_embedding_param,
update_embedding_func=params.set_embedding_param,
)
# call optimizer_wrapper.apply_gradients asynchronously
def _apply_gradients(opt_wrapper, grads_and_vars):
# sleep 1s to wait that all threads are in this method call
time.sleep(1)
opt_wrapper.apply_gradients(grads_and_vars)
executor = ThreadPoolExecutor(max_workers=thread_num)
tasks = [
executor.submit(_apply_gradients, opt_wrapper, grads_and_vars)
for grads_and_vars in grads_and_vars_batches
]
_ = [task.result() for task in tasks]
# check updated results of non-embedding variables
non_embed_vars = [
var for grad, var in grads_and_vars_batches[0][:-embed_var_n]
]
for var, expected_value in zip(non_embed_vars,
expected_non_embed_values):
self.assertTrue(np.isclose(var.numpy(), expected_value).all())
# `expected_embed_values=None` means that no need to check
# embedding table
if not expected_embed_values:
return
# check updated results of embedding table
for layer, expected_values in expected_embed_values.items():
value = params.get_embedding_param(layer, range(input_dims[layer]))
self.assertTrue(
any([
np.isclose(value, expected).all()
for expected in expected_values
]))
def setUp(self):
self.name = "embedding_1"
self.dim = 10
self.initializer = "uniform"
self.table = EmbeddingTable(self.name, self.dim, self.initializer)