def CSVPrepro(
vocab_size: int,
batch_size: int = 512,
repeat_size: Optional[int] = None,
prefetch_size: int = 1,
num_parallel_calls: int = 8,
num_negatives: int = None,
):
"""CSV Preprocessing for MovieLens."""
fields = [
F.UID,
F.INPUT_POSITIVES,
F.INPUT_MASK,
F.TARGET_POSITIVES,
F.TARGET_MASK,
F.INPUT_POSITIVES_ONE_HOT(vocab_size),
F.TARGET_POSITIVES_ONE_HOT(vocab_size),
]
return deepr.prepros.Serial(
deepr.prepros.Map(
SequenceMask(inputs="inputPositives", outputs="inputMask")),
deepr.prepros.Map(
SequenceMask(inputs="targetPositives", outputs="targetMask")),
deepr.prepros.PaddedBatch(batch_size=batch_size, fields=fields),
deepr.prepros.Map(
RandomNegatives(inputs="targetPositives",
outputs="targetNegatives",
num_negatives=num_negatives,
vocab_size=vocab_size))
if num_negatives is not None else [],
deepr.prepros.Repeat(repeat_size, modes=[deepr.TRAIN]),
deepr.prepros.Prefetch(prefetch_size),
num_parallel_calls=num_parallel_calls,
)
def __init__(self, path_csv_tr: str, path_csv_te: str, vocab_size: int):
self.path_csv_tr = path_csv_tr
self.path_csv_te = path_csv_te
self.vocab_size = vocab_size
self.fields = [
fields.UID,
fields.INPUT_POSITIVES,
fields.TARGET_POSITIVES,
fields.INPUT_POSITIVES_ONE_HOT(vocab_size),
fields.TARGET_POSITIVES_ONE_HOT(vocab_size),
]
def __init__(self,
path_csv: str,
vocab_size: int,
target_ratio: float = None,
shuffle: bool = True,
seed: int = 98765):
self.path_csv = path_csv
self.vocab_size = vocab_size
self.target_ratio = target_ratio
self.shuffle = shuffle
self.seed = seed
self.fields = [
fields.UID,
fields.INPUT_POSITIVES,
fields.TARGET_POSITIVES,
fields.INPUT_POSITIVES_ONE_HOT(vocab_size),
fields.TARGET_POSITIVES_ONE_HOT(vocab_size),
]