AbstractSuggester Documentation

Overview

AbstractSuggester is an abstract base class that defines the interface for all suggester implementations in the embedding studio. It establishes a contract by requiring the implementation of the get_topk_suggestions method and the phrases_manager property. The class standardizes the suggestion workflow by outlining how to generate suggestions based on input requests, allowing for flexibility and customization by subclasses.

Method: get_topk_suggestions

Functionality

This method returns the top-k suggestion candidates using the provided request. It retrieves candidate phrases from the phrases_manager, computes scores, and returns the highest ranking suggestions as a list of Suggest objects.

Parameters

• request: A SuggestingRequest instance containing details such as the text context for suggestions.
• top_k: An integer specifying the number of top suggestions to return. Defaults to 10.

Return Value

A list of Suggest objects, each encapsulating a suggestion text and its associated score.

Implementation Example

Subclasses must override this abstract method. An example implementation is shown below:

class MySuggester(AbstractSuggester):
    def get_topk_suggestions(self, request: SuggestingRequest, top_k: int = 10) -> List[Suggest]:
        candidate_phrases = self.phrases_manager.get_candidate_phrases(request.text)
        scored_phrases = [(self._score_phrase(phrase, request), phrase) for phrase in candidate_phrases]
        scored_phrases.sort(reverse=True)
        results = [Suggest(text=phrase.text, score=score) for score, phrase in scored_phrases[:top_k]]
        return results

Purpose

This method defines a contract for obtaining text suggestions based on the context provided in a SuggestingRequest. It allows flexibility in how suggestions are scored and returned, enabling custom implementation by subclasses without altering the interface.

---

Property: phrases_manager

Functionality

Returns the phrase manager instance used by the suggester. The phrase manager is responsible for managing candidate phrases for suggestions.

Parameters

None.

Usage

• Purpose: Retrieve the phrases manager in a subclass of AbstractSuggester.

Example

class MySuggester(AbstractSuggester):
    @property
    def phrases_manager(self) -> AbstractSuggestionPhraseManager:
        return self._phrase_manager

---

Inheritance

AbstractSuggester inherits from Python's ABC module and enforces that all extending classes implement its abstract methods.

Usage

Subclasses must override: - get_topk_suggestions: Returns a list of suggestions based on a request. - phrases_manager: Manages candidate phrases and scoring.

Example

Below is an example subclass implementation:

class CustomSuggester(AbstractSuggester):
    @property
    def phrases_manager(self):
        return custom_phrase_manager

    def get_topk_suggestions(self, request, top_k=10):
        candidates = self.phrases_manager.get_candidate_phrases(request.text)
        # Logic for scoring and ranking candidates.
        return candidates[:top_k]