File size: 3,337 Bytes
8f447ab
31f0e47
8f447ab
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
31f0e47
8f447ab
 
 
 
 
 
31f0e47
 
 
 
8f447ab
 
31f0e47
 
8f447ab
31f0e47
8f447ab
31f0e47
 
 
8f447ab
 
 
31f0e47
 
 
 
 
8f447ab
 
31f0e47
8f447ab
31f0e47
 
8f447ab
ac9dd65
 
 
 
 
 
 
 
 
 
 
8f447ab
31f0e47
 
 
 
 
8f447ab
31f0e47
8f447ab
31f0e47
 
8f447ab
 
31f0e47
8f447ab
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
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
"""
business_sim_env.py - Client-side wrapper for the Business Sim environment.

Mirrors the BrowserGymEnv.from_docker_image() pattern from the sample
inference script so inference.py can swap in cleanly.
"""
import httpx
from src.models import CEOAction, CompanyObservation, StepResult


class _Result:
    """Return type that mirrors BrowserGym's result objects."""
    def __init__(self, data: dict):
        self.observation = CompanyObservation(**data["observation"])
        self.reward      = data.get("reward", 0.0)
        self.done        = data.get("done", False)
        self.info        = data.get("info", {})


class _ResetResult:
    def __init__(self, obs_data: dict):
        self.observation = CompanyObservation(**obs_data)
        self.reward      = 0.0
        self.done        = False
        self.info        = {}


class BusinessSimEnv:
    """
    HTTP client that wraps the FastAPI environment server.
    Usage mirrors BrowserGymEnv so inference.py stays clean.
    """

    def __init__(
        self,
        base_url: str = "http://localhost:7860",
        task_id:  str = "single_quarter_survival",
    ):
        self.base_url   = base_url.rstrip("/")
        self.task_id    = task_id
        self.session_id: str | None = None
        self._client    = httpx.Client(timeout=30)

    # -- Mirrors BrowserGymEnv.from_docker_image() -----------------------------
    @classmethod
    def from_docker_image(
        cls,
        image:    str,
        env_vars: dict,
    ) -> "BusinessSimEnv":
        """Factory method using env vars similar to OpenEnv CI."""
        url     = env_vars.get("BUSINESS_SIM_URL", "http://localhost:7860")
        task_id = env_vars.get("BUSINESS_SIM_TASK", "single_quarter_survival")
        return cls(base_url=url, task_id=task_id)

    def reset(self) -> _ResetResult:
        """Call POST /reset?task_id=..."""
        resp = self._client.post(
            f"{self.base_url}/reset",
            params={"task_id": self.task_id}
        )
        resp.raise_for_status()
        data = resp.json()
        self.session_id = data["session_id"]
        return _ResetResult(data)

    def step(self, action: CEOAction) -> _Result:
        """Call POST /step?session_id=..."""
        if not self.session_id:
            raise RuntimeError("Must call reset() before step()")
            
        resp = self._client.post(
            f"{self.base_url}/step",
            params={"session_id": self.session_id},
            json=action.model_dump()
        )
        resp.raise_for_status()
        return _Result(resp.json())

    def get_state(self) -> dict:
        """Call GET /state?session_id=..."""
        if not self.session_id:
            return {}
        resp = self._client.get(
            f"{self.base_url}/state",
            params={"session_id": self.session_id}
        )
        resp.raise_for_status()
        return resp.json()

    def grade(self) -> float:
        """Call GET /grade?session_id=..."""
        if not self.session_id:
            return 0.5
            
        resp = self._client.get(
            f"{self.base_url}/grade",
            params={"session_id": self.session_id}
        )
        resp.raise_for_status()
        return float(resp.json()["score"])

    def close(self):
        """Cleanup."""
        self._client.close()