graph LR
eval_pos_neg["eval_pos_neg"]
eval_rank["eval_rank"]
eval_pos_neg -- "invokes" --> model_predict
eval_pos_neg -- "invokes" --> eval_rank
eval_rank -- "computes" --> metrics
The reclearn.evaluator subsystem is designed to provide a structured approach to evaluating recommendation algorithms. It primarily consists of two core components: eval_pos_neg and eval_rank. The eval_pos_neg component acts as the initial orchestrator, preparing data and leveraging a model to generate predictions. These predictions are then passed to eval_rank, which is responsible for calculating various ranking-based metrics. This clear separation of concerns ensures that data preparation and model interaction are handled distinctly from the metric computation, leading to a modular and maintainable evaluation pipeline.
This component serves as the high-level entry point and orchestrator for evaluations involving both positive and negative samples. Its primary responsibility is to prepare the input data, manage the context for metric calculations, and initiate the evaluation flow by interacting with a prediction model. It acts as a facade for the overall evaluation process, abstracting away the complexities of data preparation for different evaluation scenarios.
Related Classes/Methods:
This component specializes in computing a variety of ranking-based evaluation metrics, including Precision, Recall, Normalized Discounted Cumulative Gain (NDCG), and Mean Reciprocal Rank (MRR). It takes processed data or ranking scores as input and calculates the performance indicators relevant to ranking tasks.
Related Classes/Methods: