config.py

# ──────────────────────────────────────────────────────────
#  Project  : Snake_AI
#  File     : config.py
#  Author   : ProGen18
#  Created  : 25-02-2026
#  Modified : 27-03-2026
#  Purpose  : Store all configuration parameters for training and UI
# ──────────────────────────────────────────────────────────


# ============================================================
#                          IMPORTS
# ============================================================

from dataclasses import dataclass


# ============================================================
#                       CONFIGURATION
# ============================================================

# ╔══════════════════════════════════════════════════════════╗
# ║                   ConfigEntrainement                     ║
# ╠══════════════════════════════════════════════════════════╣
# ║  Handles:                                                ║
# ║   • Training hyperparameters                             ║
# ║   • Environment dimensions and limits                    ║
# ║   • Neural network architecture sizing                   ║
# ╚══════════════════════════════════════════════════════════╝


@dataclass
class ConfigEntrainement:
    """
    Configuration for the Snake AI training environment and model properties.

    Attributes:
        graine (int): Random seed for reproducibility.
        nb_environnements (int): Number of parallel environments.
        largeur (int): Width of the game area.
        hauteur (int): Height of the game area.
        taille_bloc (int): Size of a grid block.
        memoire_max (int): Maximum replay memory size.
    """

    graine: int = 42  # Seed for random generation
    nb_environnements: int = 50  # Number of parallel envs
    largeur: int = 640  # Width of the game area
    hauteur: int = 480  # Height of the game area
    taille_bloc: int = 20  # Size of a grid block
    memoire_max: int = 200000  # Max replay memory size
    taille_batch: int = 512  # Main batch size
    taille_batch_pqn: int = 512  # PQN specific batch size
    memoire_pommes_capacite: int = 20000  # Apple memory capacity
    memoire_pommes_seuil: int = 64  # Apple memory threshold
    memoire_pommes_echantillons: int = 128  # Apple memory samples
    input_size: int = 26  # Model input size
    output_size: int = 3  # Model output size
    taille_couche_1: int = 256  # First hidden layer size
    taille_couche_2: int = 256  # Second hidden layer size
    taille_couche_v: int = 128  # Value stream layer size
    taille_couche_a: int = 128  # Advantage stream layer size
    taux_apprentissage: float = 0.0003  # Learning rate
    gamma: float = 0.99  # Discount factor
    tau: float = 0.005  # Soft update rate
    epsilon_depart: float = 1.0  # Initial exploration rate
    epsilon_fin: float = 0.02  # Final exploration rate
    epsilon_frames: int = 10000  # Frames to decay epsilon
    blend_frames: int = 5000  # Frames for blending
    freq_entrainement: int = 8  # Training frequency
    n_step: int = 3  # N-step return window
    transitions_min_debut: int = 5000  # Min transitions before training
    eval_intervalle: int = 5000  # Evaluation interval
    nb_episodes_test: int = 100  # Number of test episodes
    famine_base: int = 100  # Base starvation limit
    famine_par_case: int = 5  # Extra starvation per block
    recompense_step: float = -0.001  # Step penalty
    recompense_pomme: float = 1.0  # Apple reward
    recompense_mort: float = -1.0  # Death penalty
    flood_depth_facteur: float = 1  # Flood fill depth factor
    flood_depth_max: int = 50  # Flood fill max depth
    densite_rayon: int = 5  # Density ray size
    scores_historique_maxlen: int = 500  # Max history length for scores
    scores_test_maxlen: int = 100  # Max test history length
    log_intervalle_sec: float = 1.0  # Log interval in seconds
    graph_update_intervalle: int = 100  # Graph update interval
    lr_scheduler_epsilon_seuil: float = 0.2  # Epsilon threshold for LR scheduler
    lr_scheduler_patience: int = 100  # LR scheduler patience
    lr_scheduler_factor: float = 0.5  # LR scheduler factor
    lr_min: float = 1e-06  # Minimum learning rate


# ╔══════════════════════════════════════════════════════════╗
# ║                   ConfigAffichage                        ║
# ╠══════════════════════════════════════════════════════════╣
# ║  Handles:                                                ║
# ║   • Display settings and dimensions                      ║
# ║   • Screenshot capture settings                          ║
# ╚══════════════════════════════════════════════════════════╝


@dataclass
class ConfigAffichage:
    """
    Configuration for the application's UI display parameters.

    Attributes:
        largeur_fenetre (int): Dashboard total width.
        hauteur_fenetre (int): Dashboard total height.
        largeur_menu (int): Menu panel width.
        hauteur_barre_bas (int): Bottom bar height.
        dossier_screenshots (str): Folder for screenshots.
        intervalle_screenshot_defaut (int): Default screenshot interval.
    """

    largeur_fenetre: int = 1920  # Window width
    hauteur_fenetre: int = 1080  # Window height
    largeur_menu: int = 250  # Menu width
    hauteur_barre_bas: int = 40  # Bottom bar height
    dossier_screenshots: str = "screenshots"  # Screenshot output folder
    intervalle_screenshot_defaut: int = 60  # Screenshot capture interval


# ============================================================
#                      GLOBAL INSTANCES
# ============================================================

CONFIG = ConfigEntrainement()
CONFIG_UI = ConfigAffichage()