File size: 2,357 Bytes
b94ec37
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4b7241b
b94ec37
 
4b7241b
b94ec37
 
 
 
 
4b7241b
b94ec37
 
4b7241b
 
 
 
 
b94ec37
4b7241b
 
 
 
b94ec37
 
4b7241b
 
 
 
 
 
 
 
 
 
b94ec37
4b7241b
b94ec37
 
 
 
 
 
 
 
4b7241b
b94ec37
 
 
 
 
 
 
4b7241b
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
73
74
75
76
77
78
79
80
81
82
83
84
85
import random
from .models import Observation, Action
from .tasks import TASKS
from .grader import grade

class CustomerSupportEnv:

    def __init__(self):
        self.task = None
        self.history = []
        self.steps = 0
        self.status = "open"
        self.progress = 0.0

    def reset(self):
        self.task = random.choice(list(TASKS.values()))
        self.history = []
        self.steps = 0
        self.status = "open"
        self.progress = 0.0

        return Observation(
            customer_query=self.task["query"],
            conversation_history=[],
            ticket_status=self.status,
            sentiment="negative",
            progress=0.0
        )

    def step(self, action: Action):
        self.steps += 1

        # 🎯 Get score from grader
        score, feedback = grade(action, self.task, self.steps)

        # πŸ” Loop penalty (same action repeated)
        if self.history:
            last = self.history[-1]["agent"]
            if last["action_type"] == action.action_type:
                score -= 0.1

        # πŸ“Š Progress update (smooth learning signal)
        self.progress = (self.progress + score) / 2

        # 🎯 Check if final correct action
        is_correct_action = (
            action.action_type == self.task["expected_action"] and
            action.category == self.task["expected_category"]
        )

        # 🏁 DONE LOGIC (FINAL FIX)
        if is_correct_action:
            done = True
            score = max(score, 0.85)  # ensure high reward
            self.status = "closed"

        elif self.steps >= 5:
            done = True  # stop long episodes

        else:
            done = False

        # πŸ“ Save history
        self.history.append({
            "agent": action.model_dump()
        })

        # πŸ“¦ Observation
        obs = Observation(
            customer_query=self.task["query"],
            conversation_history=self.history,
            ticket_status=self.status,
            sentiment="neutral" if done else "negative",
            progress=self.progress
        )

        return obs, max(0.0, min(1.0, score)), done, {"feedback": feedback}

    def state(self):
        return {
            "task": self.task,
            "history": self.history,
            "steps": self.steps,
            "progress": self.progress
        }