recpack.algorithms.FactorizationAlgorithm

class recpack.algorithms.FactorizationAlgorithm(num_components=100)

Base class for factorization algorithms

During fitting two matrices are constructed.

user_embedding_ contains the users embedded in a lower dimensional space, shape = |users| x num_components
item_embedding_ contains the items embedded in the same dimensions shape = num_components x |items|

Prediction happens by multiplying a user’s features with the item embedding.

Usually a child class will have to implement just the _fit() method, to construct the self.user_embedding_ and self.item_embedding_ attributes.

Parameters: num_components (int, optional) – the dimension of the feature matrices. defaults to 100

Methods

`_check_fit_complete`()	Helper function to check that fit stored information.
`_predict`(X)	Predict scores for nonzero users in the interaction matrix

Attributes

`identifier`	Name of the object.
`name`	Name of the object's class.

_check_fit_complete()

Helper function to check that fit stored information.

Checks implemented:

Checks if model is fitted, using the sklearn check_is_fitted function, https://scikit-learn.org/stable/modules/generated/sklearn.utils.validation.check_is_fitted.html
Checks if num_components is correct

_predict(X: scipy.sparse._csr.csr_matrix) → scipy.sparse._csr.csr_matrix

Predict scores for nonzero users in the interaction matrix

For each nonzero users their embedding is multiplied with the item embeddings, and scores are the result of this matrix multiplication.

Before recommending checks the shape of X to the number of items and users in the embeddings.