pasteur.transform.RefTransformer

class pasteur.transform.RefTransformer(**_)

Reference Transformers use a reference column as an input to create their embeddings.

They can be used to integrate constraints (and domain knowledge) into embeddings, in such a way that all embeddings produce valid solutions and learning is easier.

For example, consider an end date embedding that references a start date. The embedding will form a stable histogram with much less entropy, based on the period length. In addition, provided that the embedding is forced to be positive, any value it takes will produce a valid solution.

Attributes

deterministic	For a given output, the input is the same.
lossless	The decoded output equals the input.
stateful	Transformer fits variables.
name

Methods

fit(data[, ref])	Fits to the provided data
fit_transform(data[, ref])
get_attributes()
get_factory(*args, **kwargs)	Returns a factory that registers this module to the system.
reduce(other)
reverse(data[, ref])	When reversing, the data column contains encoded data, whereas the ref column contains decoded/original data.
transform(data[, ref])

deterministic = True

For a given output, the input is the same.

fit(data, ref=None)

Fits to the provided data

fit_transform(data, ref=None)

Return type:

DataFrame

get_attributes()

Return type:

Mapping[str | tuple[str, ...], Attribute]

classmethod get_factory(*args, **kwargs)

Returns a factory that registers this module to the system.

Any *args and **kwargs passed to this function will be saved and passed to the module’s __init__() method when calling build().

lossless = True

The decoded output equals the input.

name: str

reduce(other)

reverse(data, ref=None)

When reversing, the data column contains encoded data, whereas the ref column contains decoded/original data. Therefore, the referred columns have to be decoded first.

Return type:

DataFrame

stateful = False

Transformer fits variables.

transform(data, ref=None)

Return type:

DataFrame