poprox_recommender.evaluation.generate.outputs

Classes

EmbeddingWriter(outs)	Implementation of RecommendationWriter that extracts the candidate embeddings and writes them to disk.
JSONRecommendationWriter(outs)	Implementation of RecommendationWriter that writes the recommendations in compressed NDJSON format for detailed analysis.
OfflineRecResults(*, final[, ranked, reranked])
OfflineRecommendations(*, request, results)
ParquetRecommendationWriter(outs)	Implementation of RecommendationWriter that writes the recommendations in tabular format to Parquet for easy analysis.
RecOutputs(dir)	Abstraction of output layout to put it in one place.
RecommendationWriter()	Interface for recommendation writers that write various aspects of recommendations to disk.

class poprox_recommender.evaluation.generate.outputs.OfflineRecommendations(**data)

Bases: BaseModel

Parameters:

• data (Any)

• request (RecommendationRequestV1)

• results (OfflineRecResults)

class poprox_recommender.evaluation.generate.outputs.OfflineRecResults(*, final, ranked=None, reranked=None)

Bases: BaseModel

Parameters:

• final (RecommendationList)

• ranked (RecommendationList | None)

• reranked (RecommendationList | None)

class poprox_recommender.evaluation.generate.outputs.RecOutputs(dir)

Bases: object

Abstraction of output layout to put it in one place.

Parameters:

dir (Path)

property rec_dir

Output directory for recommendations. This entire directory should be read with pd.read_parquet(), and it will load the shareds from it.

property rec_parquet_file

Output file for recommendations in Parquet tabular format.

property rec_json_file

Output file for recommendations in NDJSON format.

property emb_file

File for final deduplicated embedding outputs.

class poprox_recommender.evaluation.generate.outputs.RecommendationWriter

Bases: ABC

Interface for recommendation writers that write various aspects of recommendations to disk.

classmethod make_actor()

Turn this writer into a Ray actor class.

Return type:

ActorClass

abstractmethod write_recommendations(request, pipeline_state)

Write recommendations to this writer’s storage.

Parameters:

• request (RecommendationRequestV1)

• pipeline_state (PipelineState)

class poprox_recommender.evaluation.generate.outputs.ParquetRecommendationWriter(outs)

Bases: RecommendationWriter

Implementation of RecommendationWriter that writes the recommendations in tabular format to Parquet for easy analysis.

Can be used as a Ray actor.

Parameters:

outs (RecOutputs)

write_recommendations(request, pipeline_state)

Write recommendations to this writer’s storage.

Parameters:

• request (RecommendationRequestV1)

• pipeline_state (PipelineState)

class poprox_recommender.evaluation.generate.outputs.JSONRecommendationWriter(outs)

Bases: RecommendationWriter

Implementation of RecommendationWriter that writes the recommendations in compressed NDJSON format for detailed analysis.

Can be used as a Ray actor.

Parameters:

outs (RecOutputs)

write_recommendations(request, pipeline_state)

Write recommendations to this writer’s storage.

Parameters:

• request (RecommendationRequestV1)

• pipeline_state (PipelineState)

class poprox_recommender.evaluation.generate.outputs.EmbeddingWriter(outs)

Bases: RecommendationWriter

Implementation of RecommendationWriter that extracts the candidate embeddings and writes them to disk.

Can be used as a Ray actor.

Parameters:

outs (RecOutputs)

write_recommendations(request, pipeline_state)

Write recommendations to this writer’s storage.

Parameters:

• request (RecommendationRequestV1)

• pipeline_state (PipelineState)