Config.py

import atexit
import os
import random
from pathlib import Path

import numpy as np
import torch
from dotenv import load_dotenv
from lightning import seed_everything

load_dotenv()


class _DBConfig:
    db_name = os.getenv("POSTGRES_DB")
    db_password = os.getenv("POSTGRES_PASSWORD")
    db_username = os.getenv("POSTGRES_USER")
    db_url = f"postgresql://{db_username}:{db_password}@localhost:5432/{db_name}"


class _ConfigPaths:
    root = Path(__file__).parent
    model_path = root / "models"
    model_path.mkdir(exist_ok=True)
    gui = root / "gui"
    templates = gui / "templates"
    ai_weights = root / 'speed_game.pth'
    train_values_file = (root / "train_log.log").open('w')


class _ConfigAgent:
    hidden_size = (
        256,
        128,
        64,
        32,
        32,
        32,
        32,
        32,
        32,
    )
    initial_train_learning_rate = 1e-3
    debug = True
    # debug = False
    if not debug:
        random_state = random.randint(0, 2 ** 32)
        print(f"{random_state=}")
    else:
        random_state = 42


class Config(_ConfigPaths, _ConfigAgent, _DBConfig):
    lr_decline_rate = 1000
    agent_print_interval = 10
    results_over_time_counter = 100
    agent_save_interval = 500
    play_beta = 100
    beta = 100
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    training_buffer_len = 50_000
    min_n_points_to_finish = 15
    n_players = 2
    n_actions = 45


seed_everything(Config.random_state, workers=True)
random.seed(Config.random_state)
np.random.seed(Config.random_state)


@atexit.register
def close():
    Config.train_values_file.close()