def train(store: BaseConfig,
tokenizer_trainer: Optional[BaseTokenizerTrainer] = None):
"""Train a Synthetic Model. This is a facade entrypoint that implements the engine
specific training operation based on the provided configuration.
Args:
store: A subclass instance of ``BaseConfig.`` This config is reponsible for
providing the actual training entrypoint for a specific training routine.
tokenizer_trainer: An optional subclass instance of a ``BaseTokenizerTrainer``. If provided
this tokenizer will be used to pre-process and create an annotated dataset for training.
If not provided a default tokenizer will be used.
"""
if tokenizer_trainer is None:
tokenizer_trainer = _create_default_tokenizer(store)
tokenizer_trainer.create_annotated_training_data()
tokenizer_trainer.train()
tokenizer = tokenizer_from_model_dir(store.checkpoint_dir)
params = TrainingParams(tokenizer_trainer=tokenizer_trainer,
tokenizer=tokenizer,
config=store)
train_fn = store.get_training_callable()
store.save_model_params()
store.gpu_check()
train_fn(params)
def train_rnn(store: BaseConfig):
"""
Fit synthetic data model on training data.
This will annotate the training data and create a new file that
will be used to actually train on. The updated training data, model,
checkkpoints, etc will all be saved in the location specified
by your config.
Args:
store: An instance of one of the available configs that you
previously created
Returns:
None
"""
if not store.overwrite: # pragma: no cover
try:
_load_model(store)
except Exception:
pass
else:
raise RuntimeError(
'A model already exists in the checkpoint location, you must enable overwrite mode or delete the checkpoints first.'
) # noqa
text = _annotate_training_data(store)
sp = _train_tokenizer(store)
dataset = _create_dataset(store, text, sp)
logging.info("Initializing synthetic model")
model = _build_sequential_model(vocab_size=len(sp),
batch_size=store.batch_size,
store=store)
# Save checkpoints during training
checkpoint_prefix = (Path(store.checkpoint_dir) / "synthetic").as_posix()
if store.save_all_checkpoints:
checkpoint_prefix = checkpoint_prefix + "-{epoch}"
checkpoint_callback = tf.keras.callbacks.ModelCheckpoint(
filepath=checkpoint_prefix, save_weights_only=True, monitor='accuracy')
history_callback = _LossHistory()
model.fit(dataset,
epochs=store.epochs,
callbacks=[checkpoint_callback, history_callback])
_save_history_csv(history_callback, store.checkpoint_dir)
store.save_model_params()
logging.info(
f"Saving model to {tf.train.latest_checkpoint(store.checkpoint_dir)}")
if store.dp:
logging.info(_compute_epsilon(len(text), store))
def __init__(
self,
config: BaseConfig,
*,
seed_list: List[str],
line_validator: Optional[Callable] = None,
max_invalid: int = 1000,
):
generator_class = config.get_generator_class()
tokenizer = tokenizer_from_model_dir(config.checkpoint_dir)
self.settings = Settings(
config=config,
start_string=seed_list,
line_validator=line_validator,
max_invalid=max_invalid,
tokenizer=tokenizer,
)
self._generator = generator_class(self.settings).generate_next(None)
def generate_text(config: BaseConfig,
start_string: Optional[str] = None,
line_validator: Optional[Callable] = None,
max_invalid: int = 1000,
num_lines: Optional[int] = None,
parallelism: int = 0) -> Iterator[GenText]:
"""A generator that will load a model and start creating records.
Args:
config: A configuration object, which you must have created previously
start_string: A prefix string that is used to seed the record generation.
By default we use a newline, but you may substitue any initial value here
which will influence how the generator predicts what to generate. If you
are working with a field delimiter, and you want to seed more than one column
value, then you MUST utilize the field delimiter specified in your config.
An example would be "foo,bar,baz,". Also, if using a field delimiter, the string
MUST end with the delimiter value.
line_validator: An optional callback validator function that will take
the raw string value from the generator as a single argument. This validator
can executue arbitrary code with the raw string value. The validator function
may return a bool to indicate line validity. This boolean value will be set
on the yielded ``gen_text`` object. Additionally, if the validator throws an
exception, the ``gen_text`` object will be set with a failed validation. If the
validator returns None, we will assume successful validation.
max_invalid: If using a ``line_validator``, this is the maximum number of invalid
lines to generate. If the number of invalid lines exceeds this value a ``RunTimeError``
will be raised.
num_lines: If not ``None``, this will override the ``gen_lines`` value that is provided in the ``config``
parallelism: The number of concurrent workers to use. ``1`` (the default) disables parallelization,
while a non-positive value means "number of CPUs + x" (i.e., use ``0`` for using as many workers
as there are CPUs). A floating-point value is interpreted as a fraction of the available CPUs,
rounded down.
Simple validator example::
def my_validator(raw_line: str):
parts = raw_line.split(',')
if len(parts) != 5:
raise Exception('record does not have 5 fields')
NOTE:
``gen_lines`` from the ``config`` is important for this function. If a line validator is not provided,
each line will count towards the number of total generated lines. When the total lines generated is >=
``gen_lines`` we stop. If a line validator is provided, only *valid* lines will count towards
the total number of lines generated. When the total number of valid lines generated is >= ``gen_lines``,
we stop.
NOTE:
``gen_chars``, controls the possible maximum number of characters a single
generated line can have. If a newline character has not been generated before reaching
this number, then the line will be returned. For example if ``gen_chars`` is 180 and a
newline has not been generated, once 180 chars have been created, the line will be returned
no matter what. As a note, if this value is 0, then each line will generate until
a newline is observed.
Yields:
A ``GenText`` object for each record that is generated. The generator
will stop after the max number of lines is reached (based on your config).
Raises:
A ``RunTimeError`` if the ``max_invalid`` number of lines is generated
"""
generator_class = config.get_generator_class()
tokenizer = tokenizer_from_model_dir(config.checkpoint_dir)
settings = Settings(config=config,
start_string=start_string,
line_validator=line_validator,
max_invalid=max_invalid,
generator=generator_class,
tokenizer=tokenizer)
if num_lines is not None:
_line_count = num_lines
else:
_line_count = config.gen_lines
num_workers = get_num_workers(parallelism, _line_count, chunk_size=5)
if num_workers == 1:
gen = generator_class(settings)
yield from gen.generate_next(_line_count)
else:
yield from generate_parallel(settings,
_line_count,
num_workers,
chunk_size=5)