-
Notifications
You must be signed in to change notification settings - Fork 13
/
evaluator_greedy.py
72 lines (54 loc) · 2.29 KB
/
evaluator_greedy.py
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
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
import numpy as np
# Create the evaluation environment
from env_creator import qsimpy_env_creator
eval_env = qsimpy_env_creator(
env_config={
"obs_filter": "rescale_-1_1",
"reward_filter": None,
"dataset": "qdataset/traintest/synds_200_sub_60_testset_with_sorted_arrival.csv",
"config": {"evaluation": True, "policy": "greedy"},
}
)
skipped_list = []
def enhanced_greedy_policy(env, skipped_list):
current_obs = env.current_obs
qnode_start_index = 4 # Adjust based on the actual qtask observation length
qnode_obs_length = 3 # Number of values per qnode in the observation
best_choice = None
earliest_time = float("inf")
# Iterate over each qnode observation to find the one with the earliest next available time
for i in range(qnode_start_index, len(current_obs), qnode_obs_length):
qnode_id = (i - qnode_start_index) // qnode_obs_length
# Skip the qnode if it's in the skipped list
if qnode_id in skipped_list:
continue
qnode_next_available_time = current_obs[i + 2] # next_available_time of qnode
# Update the best choice if this qnode's next available time is earlier
if qnode_next_available_time < earliest_time:
earliest_time = qnode_next_available_time
best_choice = qnode_id
return best_choice
# Run the evaluation
num_episodes = 1 # Set the number of episodes for evaluation
for episode in range(num_episodes):
obs = eval_env.reset(seed=22)
# print(f"Episode {episode} started with observation: {obs}")
terminate = False
episode_reward = 0
while not terminate:
action = enhanced_greedy_policy(eval_env, skipped_list)
obs, reward, terminate, _, info = eval_env.step(action)
episode_reward += reward
# If the reward is -10, add the action (qnode_id) to the skipped list
if reward <= -10:
skipped_list.append(action)
# print(
# f"Action: {action}, Reward: {reward}, Terminate: {terminate}, Skip list: {skipped_list}"
# )
if reward > -10:
skipped_list.clear()
if terminate:
print(f"Episode {episode} finished with reward {episode_reward}")
break
# Run and close the evaluation environment
eval_env.close()