Response functions

Response functions are used to model users behaviour on items to simulate any kind of reaction that user perform. For example it can be binary classification model that determines whether user clicked on a item, or rating, that user gave to an item.

Base classes

All of the existing response functions are made on a top of underlying base classes ActionModelEstimator and ActionModelTransformer which follows the logic of spark’s Estimator and Transformer classes. To implement custom response function the one can inherit from base classes:

  • ActionModelEstimator if any learning logic is necessary

  • ActionModelTransformer for performing model infering

Base classes are inherited from spark’s Estimator and Transformer and to define fit() or transform() logic the one should overwrite _fit() and _transform() respectively. Note, that those base classes are useful to implement your own response function, but are not necessary, and to create a response pipeline any proper spark’s estimators/transformers can be used

class sim4rec.response.ActionModelEstimator(outputCol: str = None)

Base class for response estimator

Parameters:

outputCol – Name of the response score column, defaults to None

class sim4rec.response.ActionModelTransformer(outputCol: str = None)

Base class for response transformer. transform() will be used to calculate score based on inputCols, and write it to outputCol column

Parameters:

outputCol – Name of the response score column, defaults to None

Response functions

class sim4rec.response.ConstantResponse(value: float = 0.0, outputCol: str | None = None)

Always returns constant valued response

Parameters:

  • value – Response value

  • outputCol – Output column name

class sim4rec.response.NoiseResponse(mu: float | None = None, sigma: float | None = None, outputCol: str | None = None, clipNegative: bool = True, seed: int | None = None)

Creates random response sampled from normal distribution

Parameters:

  • mu – Mean parameter of normal distribution

  • sigma – Standard deviation parameter of normal distribution

  • outputCol – Output column name

  • clip_negative – Whether to make response non-negative, defaults to True

  • seed – Random state seed, defaults to None

class sim4rec.response.CosineSimilatiry(inputCols: List[str] | None = None, outputCol: str | None = None)

Calculates the cosine similarity between two vectors. The result is in [0; 1] range

Parameters:

  • inputCols – Two column names with dense vectors

  • outputCol – Output column name

class sim4rec.response.BernoulliResponse(inputCol: str | None = None, outputCol: str | None = None, seed: int | None = None)

Samples responses from probability column

Parameters:

  • inputCol – Probability column name. Probability should be in range [0; 1]

  • outputCol – Output column name

  • seed – Random state seed, defaults to None

class sim4rec.response.ParametricResponseFunction(inputCols: List[str] | None = None, outputCol: str | None = None, weights: Iterable | None = None)

Calculates response based on the weighted sum of input responses

Parameters:

  • inputCols – Input responses column names

  • outputCol – Output column name

  • weights – Input responses weights