class UserCBCosineKNNRecommender(BaseUserColdStartRecommender):
def __init__(
self,
X_interaction: InteractionMatrix,
X_profile: ProfileMatrix,
top_k: int = 100,
n_threads: Optional[int] = None,
shrink: float = 1e-1,
):
self.top_k = top_k
self.shrink = shrink
self.n_threads = get_n_threads(n_threads)
super().__init__(X_interaction, X_profile)
self.X_interaction_csc = X_interaction.astype(np.float64).tocsc()
self.sim_computer: Optional[CosineSimilarityComputer] = None
def _learn(self) -> None:
self.sim_computer = CosineSimilarityComputer(self.X_profile,
self.shrink,
normalize=True,
n_threads=self.n_threads)
def get_score(self, profile: ProfileMatrix) -> DenseScoreArray:
if self.sim_computer is None:
raise RuntimeError("'get_score' called before learn.")
if not sps.issparse(profile):
profile = sps.csr_matrix(profile)
similarity = self.sim_computer.compute_similarity(profile, self.top_k)
score = sparse_mm_threaded(similarity, self.X_interaction_csc,
self.n_threads)
return score.astype(np.float64)
def _create_computer(self, X: InteractionMatrix) -> CosineSimilarityComputer:
return CosineSimilarityComputer(
X, self.shrinkage, self.normalize, self.n_threads
)
def _learn(self) -> None:
self.sim_computer = CosineSimilarityComputer(self.X_profile,
self.shrink,
normalize=True,
n_threads=self.n_threads)