game_visualization.py

# script to visualize how agent plays a game
# useful to study different iterations

import numpy as np
from agent import DeepQLearningAgent
from game_environment import Snake
from utils import visualize_game
import json

# Using the same version as in training
version = 'v17.1'

with open('model_config/{:s}.json'.format(version), 'r') as f:
    m = json.loads(f.read())
    board_size = m['board_size']
    frames = m['frames'] # keep frames >= 2
    max_time_limit = m['max_time_limit']
    supervised = bool(m['supervised'])
    n_actions = m['n_actions']
    obstacles = bool(m['obstacles'])

iteration_list = [107500, 119500]
max_time_limit = 398

# Initialize environment
env = Snake(board_size=board_size, frames=frames, max_time_limit=max_time_limit,
            obstacles=obstacles, version=version)
s = env.reset()
n_actions = env.get_num_actions()

# Initialize the Agent
agent = DeepQLearningAgent(board_size=board_size, frames=frames, 
                           n_actions=n_actions, buffer_size=10, version=version)

# Visualize the agent playing the game with weights from listed iterations
for iteration in iteration_list:
    agent.load_model(file_path='models/{:s}'.format(version), iteration=iteration)
    
    for i in range(5):
        visualize_game(env, agent,
            path='images/game_visual_{:s}_{:d}_14_ob_{:d}.mp4'.format(version, iteration, i),
            debug=False, animate=True, fps=12)