Module code_gan 

Generative Adversarial Network for Code Generation

Implements a GAN architecture for generating valid Rust AST candidates:

• Generator: Maps latent vectors to Rust AST token sequences
• Discriminator: Classifies code as real (valid) or fake (invalid)

Architecture

Latent Vector z ─┬─► Generator ─► AST Tokens ─┬─► Discriminator ─► Valid/Invalid
                 │                            │
                 │   Real AST Samples ────────┘
                 │
                 └── (sampled from N(0, I))

Example

use entrenar::generative::{CodeGan, CodeGanConfig};

let config = CodeGanConfig::default();
let mut gan = CodeGan::new(config);

// Training loop would alternate between generator and discriminator updates

Structs

CodeGan

Complete Code GAN for generating Rust AST

CodeGanConfig

Configuration for the complete Code GAN

CodeGanStats

Statistics from GAN training

Discriminator

Discriminator network: classifies code as real or fake

DiscriminatorConfig

Configuration for the Discriminator network

Generator

Generator network: maps latent vectors to AST token sequences

GeneratorConfig

Configuration for the Generator network

LatentCode

Latent code representation (vector in latent space)

TrainingResult

Training result from a GAN update step

Functions

sigmoid

Sigmoid activation function