from stable_baselines3 import DQN
from stable_baselines3.common.evaluation import evaluate_policy
from stable_baselines3.common.monitor import Monitor
import gymnasium as gym

import argparse

MODEL_NAME = "ALE-Pacman-v5"
loaded_model = DQN.load(MODEL_NAME)

# This script should have some options
# 1. Turn off the stochasticity as determined by the ALEv5
#   Even if deterministic is set to true in evaluate policy, the environment will ignore this 25% of the time
#   To compensate for this, we can set the repeat action probability to 0
#       DONE
# 2. Print out the evaluation metrics or save to file
# 3. Render in the ALE or not 
#       DONE
# 4. Print the keyword args for the environment? I think this might be helpful...
#       IN PROGRESS
# 5. 

parser = argparse.ArgumentParser()
parser.add_argument("-r", "--repeat_action_probability", help="repeat action probability", type=float, default=0.25)
parser.add_argument("-f", "--frameskip", help="frameskip", type=int, default=4)
parser.add_argument("-o", "--observe", help="observe agent", action="store_const", const=True)
parser.add_argument("-p", "--print", help="print environment information", action="store_const", const=True)
args = parser.parse_args()

# Toggle the render mode based on the -o flag
if args.observe == True:
    mode = "human"
else:
    mode = "rgb_array"

# Retrieve the environment
eval_env = Monitor(gym.make("ALE/Pacman-v5", 
                            render_mode=mode, 
                            repeat_action_probability=args.repeat_action_probability,
                            frameskip=args.frameskip,))

if args.print == True:
    env_info = str(eval_env.spec).split(", ")
    for item in env_info:
        print(item)
# Evaluate the policy
mean_rwd, std_rwd = evaluate_policy(loaded_model.policy, eval_env, n_eval_episodes=1)
print("mean rwd: ", mean_rwd)
print("std rwd: ", std_rwd)