ArKanjo: EmbeddingMethod Class Reference

Semantic duplication detector using LLM embeddings. More...

#include <embedding_method.hpp>

Inheritance diagram for EmbeddingMethod:

Collaboration diagram for EmbeddingMethod:

Public Member Functions
	EmbeddingMethod (const fs::path &base_path_, double similarity_, std::vector< std::string > pass_through_args_={})
	Constructs the LLM detector. More...

void	on_function (const FunctionData &fd) override

void	execute () override

Public Member Functions inherited from IMethod
virtual	~IMethod ()=default

virtual void	save_duplications (std::vector< DuplicationEntry > &)
	Output line format: More...

virtual void	on_function (const FunctionData &fd)=0

virtual void	execute ()=0

Detailed Description

Semantic duplication detector using LLM embeddings.

Materializes each function body as a file under the source feature directory (same as the other methods), then invokes a Python script that embeds every function with a Hugging Face model and prints the pairwise similarities, which are parsed back into duplication entries.

Definition at line 29 of file embedding_method.hpp.

Constructor & Destructor Documentation

◆ EmbeddingMethod()

EmbeddingMethod::EmbeddingMethod	(	const fs::path &	base_path_,
		double	similarity_,
		std::vector< std::string >	pass_through_args_ = `{}`
	)

Constructs the LLM detector.

Parameters

base_path_	Root path of the codebase cache.
similarity_	Minimum similarity threshold (0-100) to consider as duplicate.
pass_through_args_	Raw arguments forwarded to the embedding script (e.g. –max-seq-length, –batch-size, –model).

Definition at line 17 of file embedding_method.cpp.

Member Function Documentation

◆ execute()

void EmbeddingMethod::execute ( )

overridevirtual

Implements IMethod.

Definition at line 116 of file embedding_method.cpp.

◆ on_function()

void EmbeddingMethod::on_function ( const FunctionData & fd )