File size: 2,359 Bytes
ebb75df
 
 
 
 
c37ff18
ebb75df
cbe0575
c37ff18
 
 
 
cbe0575
 
 
 
 
 
a30d4ce
1427095
a30d4ce
1427095
c37ff18
 
a30d4ce
 
cbe0575
 
a30d4ce
 
c37ff18
ebb75df
a30d4ce
 
 
cbe0575
 
 
 
 
ebb75df
 
cbe0575
 
 
 
ebb75df
cbe0575
 
 
 
ebb75df
a30d4ce
 
c37ff18
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
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


# 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. Add option flag to accept file path for model
#       DONE
# 6. Add option flag to accept number of episodes
#       DONE

parser = argparse.ArgumentParser()
parser.add_argument("-r", "--repeat_action_probability", help="repeat action probability, default 0.25", type=float, default=0.25)
parser.add_argument("-f", "--frameskip", help="frameskip, default 4", 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)
parser.add_argument("-e", "--num_episodes", help="specify the number of episodes to evaluate, default 1", type=int, default=1)
parser.add_argument("-a", "--agent_filepath", help="file path to agent to watch, minus the .zip extension", type=str, required=True)
args = parser.parse_args()

MODEL_NAME = args.agent_filepath
loaded_model = DQN.load(MODEL_NAME)

# 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=args.num_episodes)
print("eval episodes: ", args.num_episodes)
print("mean rwd: ", mean_rwd)
print("std rwd: ", std_rwd)