Abstract base class for converting between text and integers.

A note on padding:

Because text data is typically variable length and nearly always requires padding during training, ID 0 is always reserved for padding. To accommodate this, all TextEncoders behave in certain ways:

• encode: never returns id 0 (all ids are 1+)
• decode: drops 0 in the input ids

• vocab_size: includes ID 0

  New subclasses should be careful to match this behavior.

Methods

decode

View source

@abc.abstractmethod
decode(
    ids
)

Decodes a list of integers into text.

encode

View source

@abc.abstractmethod
encode(
    s
)

Encodes text into a list of integers.

load_from_file

View source

@classmethod
@abc.abstractmethod
load_from_file(
    filename_prefix
)

Load from file. Inverse of save_to_file.

save_to_file

View source

@abc.abstractmethod
save_to_file(
    filename_prefix
)

Store to file. Inverse of load_from_file.