Spaces:
Runtime error
Runtime error
File size: 19,355 Bytes
a663682 87600b0 a663682 cbc259d a663682 cbc259d a663682 cbc259d a663682 87600b0 a663682 87600b0 a663682 87600b0 a663682 87600b0 a663682 87600b0 a663682 87600b0 a663682 87600b0 a663682 87600b0 a663682 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 | """
Cashflow Multi-Agent RL Environment.
Follows the strict 11-step workflow:
1. RESET β generate cash, invoices, revenue
2. WORLD MODEL β initialize hidden dynamics
3. AGENTS OBSERVE β partial views
4. MULTI-AGENT INTERACTION β advisors produce memos
5. CFO AGENT β resolves conflict, chooses action
6. IF NEGOTIATE β vendor agent responds
7. ENVIRONMENT STEP β update cash, invoices, credit
8. WORLD MODEL UPDATE β trigger probabilistic events
9. REWARD COMPUTATION β penalties, interest, liquidity, negotiation
10. STORE TRANSITION β (state, action, reward, reasoning)
11. LOOP β until episode ends
OpenEnv Interface: reset(), step(), state
"""
import random
from uuid import uuid4
from typing import List, Dict, Any, Optional
from openenv.core.env_server.interfaces import Environment
from openenv.core.env_server.types import State
try:
from models import (
CashflowmanagerAction, CashflowmanagerObservation,
Invoice, Receivable, VendorProfile, NegotiationResult, Transition
)
from server.world_model import WorldModel
from server.data_generator import generate_scenario
from server.agents import (
expenditure_agent, revenue_agent, risk_agent, risk_agent_icl, vendor_agent, format_memo
)
from server.reward import compute_reward
except ImportError:
try:
from cashflowmanager.models import (
CashflowmanagerAction, CashflowmanagerObservation,
Invoice, Receivable, VendorProfile, NegotiationResult, Transition
)
from cashflowmanager.server.world_model import WorldModel
from cashflowmanager.server.data_generator import generate_scenario
from cashflowmanager.server.agents import (
expenditure_agent, revenue_agent, risk_agent, risk_agent_icl, vendor_agent, format_memo
)
from cashflowmanager.server.reward import compute_reward
except ImportError:
from ..models import (
CashflowmanagerAction, CashflowmanagerObservation,
Invoice, Receivable, VendorProfile, NegotiationResult, Transition
)
from .world_model import WorldModel
from .data_generator import generate_scenario
from .agents import (
expenditure_agent, revenue_agent, risk_agent, risk_agent_icl, vendor_agent, format_memo
)
from .reward import compute_reward
MAX_DAYS = 10
class CashflowmanagerEnvironment(Environment):
"""
Multi-Agent Cashflow RL Environment with Hidden World Dynamics.
Themes:
- #1 Multi-Agent Interactions: CFO + Expenditure + Revenue + Risk + Vendor
- #3.1 World Modeling: Hidden events, partial observability
OpenEnv API:
- reset() β CashflowmanagerObservation
- step(action) β CashflowmanagerObservation
- state β State
"""
SUPPORTS_CONCURRENT_SESSIONS: bool = True
def __init__(self, difficulty="medium", seed=None):
self.difficulty = difficulty
self.seed = seed
self._state = State(episode_id=str(uuid4()), step_count=0)
self.world_model = WorldModel()
self.transitions: List[Transition] = []
self.done = False
# These get populated on reset()
self.cash = 0.0
self.credit_used = 0.0
self.credit_limit = 5000.0
self.invoices: List[Invoice] = []
self.receivables: List[Receivable] = []
self.vendors: Dict[str, VendorProfile] = {}
self.day = 1
self.last_negotiation: Optional[NegotiationResult] = None
self.last_world_events: List[str] = []
# βββββββββββββββββββββββββββββββββββββββββββββββββββ
# STEP 1: RESET β Generate cash, invoices, revenue
# βββββββββββββββββββββββββββββββββββββββββββββββββββ
def reset(self, difficulty="medium", seed=None) -> CashflowmanagerObservation:
if difficulty:
self.difficulty = difficulty
print(f"--- RESETTING ENVIRONMENT | DIFFICULTY: {self.difficulty} | SEED: {seed} ---")
if seed is not None:
random.seed(seed)
# Generate synthetic scenario based on difficulty
scenario = generate_scenario(difficulty=self.difficulty)
self.day = 1
self.cash = scenario["company"]["starting_cash"]
self.credit_used = 0.0
self.credit_limit = scenario["company"]["credit_limit"]
self.done = False
self.vendors = {v["id"]: VendorProfile(**v) for v in scenario["vendors"]}
self.invoices = [Invoice(**i) for i in scenario["initial_invoices"]]
self.receivables = [Receivable(**r) for r in scenario["initial_receivables"]]
self._state = State(episode_id=str(uuid4()), step_count=0)
self.transitions = []
self.last_negotiation = None
self.last_world_events = []
# STEP 2: WORLD MODEL β initialize hidden dynamics
self.world_model.initialize(scenario, max_days=MAX_DAYS)
# STEPS 3-4: Agents observe and produce initial memos
advisor_memos, advisor_messages = self._run_advisors()
return self._build_obs(
reward=0.0,
advisor_memos=advisor_memos,
advisor_messages=advisor_messages,
)
# βββββββββββββββββββββββββββββββββββββββββββββββββββ
# STEPS 5-10: STEP β Process action, update world, compute reward
# βββββββββββββββββββββββββββββββββββββββββββββββββββ
def step(self, action: CashflowmanagerAction) -> CashflowmanagerObservation:
self._state.step_count += 1
# Capture pre-state for transition logging
pre_state = f"Day:{self.day} Cash:{self.cash:.0f} Credit:{self.credit_used:.0f} Invoices:{len(self.invoices)}"
paid = 0
late_fee_total = 0.0
interest_total = 0.0
trust_change = 0.0
negotiation_success = False
shock_absorbed = False
self.last_negotiation = None
self.last_world_events = []
# ββ STEP 5: CFO action is received ββ
# ββ STEP 6: If NEGOTIATE β vendor responds ββ
# ββ STEP 7: Update cash, invoices, credit ββ
if action.type == "pay" and action.invoice_id:
inv = self._find_invoice(action.invoice_id)
if inv and inv.status != "paid":
pay_amount = action.amount if action.amount and action.amount > 0 else inv.amount
pay_amount = min(pay_amount, inv.amount)
available = self.cash + (self.credit_limit - self.credit_used)
pay_amount = min(pay_amount, available)
if pay_amount >= inv.amount:
inv.status = "paid"
inv.amount = 0
paid += 1
trust_change += 0.05
else:
inv.amount -= pay_amount
inv.status = "partial"
self._update_cash(-pay_amount)
elif action.type == "partial" and action.invoice_id:
inv = self._find_invoice(action.invoice_id)
if inv and inv.status != "paid":
pay_amount = action.amount if action.amount and action.amount > 0 else inv.min_payment
pay_amount = min(pay_amount, inv.amount)
available = self.cash + (self.credit_limit - self.credit_used)
pay_amount = min(pay_amount, available)
inv.amount -= pay_amount
if inv.amount <= 0:
inv.amount = 0
inv.status = "paid"
paid += 1
else:
inv.status = "partial"
self._update_cash(-pay_amount)
elif action.type == "negotiate" and action.invoice_id:
inv = self._find_invoice(action.invoice_id)
if inv and inv.status != "paid":
vendor = self.vendors.get(inv.vendor_id)
if vendor:
# STEP 6: Vendor Agent responds with mood awareness
mood = self.world_model.vendor_mood.get(inv.vendor_id, 0.0)
neg_result = vendor_agent(
{"id": vendor.id, "name": vendor.name},
inv,
vendor_mood=mood,
trust_score=vendor.trust_score
)
self.last_negotiation = {
"accepted": neg_result["accepted"],
"message": neg_result["vendor_message"]
}
if neg_result["accepted"]:
inv.due_in += neg_result["extension_days"]
negotiation_success = True
trust_change += 0.02
else:
trust_change -= 0.05 # Trust drops on rejection
# Execute Payment (Good faith payment or forced payment)
pay_amount = action.amount if action.amount and action.amount > 0 else 0.0
available = self.cash + (self.credit_limit - self.credit_used)
pay_amount = min(pay_amount, inv.amount, available)
if pay_amount > 0:
inv.amount -= pay_amount
self._update_cash(-pay_amount)
if inv.amount <= 0:
inv.amount = 0
inv.status = "paid"
paid += 1
else:
inv.status = "partial"
elif action.type == "credit":
draw_amount = action.amount if action.amount and action.amount > 0 else 100000.0
remaining_credit = self.credit_limit - self.credit_used
draw_amount = min(draw_amount, remaining_credit)
self.cash += draw_amount
self.credit_used += draw_amount
elif action.type == "defer":
pass # Explicit skip β no action taken
# Update vendor trust scores
for vid, delta in [(v, trust_change) for v in self.vendors]:
vendor = self.vendors[vid]
vendor.trust_score = max(0.0, min(1.0, vendor.trust_score + trust_change / len(self.vendors)))
# ββ STEP 8: WORLD MODEL UPDATE ββ
day_ended = self._check_day_end()
if day_ended:
# World model triggers hidden events for the day that just ended
effects = self.world_model.update(self.day)
# Apply cash shocks
if effects["cash_delta"] != 0:
self._update_cash(effects["cash_delta"])
if effects["shock_occurred"] and self.cash >= 0:
shock_absorbed = True
# Log all triggered world events
for ev in effects["events_triggered"]:
if ev["type"] in ["cash_shock", "fraud"]:
self.last_world_events.append(f"π₯ {ev['description']}")
elif ev["type"] == "revenue_miss":
self.last_world_events.append(f"π {ev['description']}")
elif ev["type"] == "vendor_shift":
self.last_world_events.append(f"π€ {ev['description']}")
# Age all unpaid invoices
for inv in self.invoices:
if inv.status != "paid":
inv.due_in -= 1
if inv.due_in < 0 and inv.amount > 0:
late_fee_total += inv.late_fee
inv.amount += inv.late_fee
if inv.amount > 0:
interest = inv.amount * inv.interest
interest_total += interest
inv.amount += interest
self.day += 1
# Apply payment delays (Revenue Delay -> Expected Revenue = 0)
for rec_id, extra_days in effects["payment_delays"]:
rec = next((r for r in self.receivables if r.id == rec_id), None)
if rec:
rec.amount = 0
rec.probability = 0
self.last_world_events.append(f"β οΈ Revenue Delay: Payment from {rec.customer_id} failed (Expected Revenue = 0)")
self.receivables.remove(rec)
# Apply vendor trust changes
for vid, delta in effects["vendor_trust_deltas"].items():
if vid in self.vendors:
self.vendors[vid].trust_score = max(0.0, min(1.0, self.vendors[vid].trust_score + delta))
# Process receivables
for rec in list(self.receivables):
rec.expected_in -= 1
if rec.expected_in <= 0:
if random.random() < rec.probability:
self.cash += rec.amount
self.last_world_events.append(f"π° Received ${rec.amount:.0f} from {rec.customer_id}")
else:
self.last_world_events.append(f"β Payment from {rec.customer_id} defaulted")
self.receivables.remove(rec)
# Check for fraud
if effects["fraud_alert"]:
self.last_world_events.append("π¨ Fraud alert β investigation triggered")
if effects["revenue_miss"]:
self.last_world_events.append("π Revenue target missed β board review")
# Episode termination
if self.day > MAX_DAYS:
self.done = True
# Check for bankruptcy before reward
if self.cash < -500000: # Threshold for total collapse
self.done = True
# ββ STEP 9: REWARD COMPUTATION ββ
reward = compute_reward(
cash=self.cash,
late_fee=late_fee_total,
interest=interest_total,
credit_used=self.credit_used,
credit_limit=self.credit_limit,
paid=paid,
is_bankrupt=self.done,
negotiation_success=negotiation_success
)
# ββ STEPS 3-4: Advisors observe new state and produce memos ββ
advisor_memos, advisor_messages = self._run_advisors()
# ββ STEP 10: STORE TRANSITION ββ
post_state = f"Day:{self.day} Cash:{self.cash:.0f} Credit:{self.credit_used:.0f} Invoices:{len([i for i in self.invoices if i.status != 'paid'])}"
self.transitions.append(Transition(
day=self.day,
state_summary=pre_state,
advisor_memos=advisor_memos,
action={"type": action.type, "invoice_id": action.invoice_id, "amount": action.amount},
reward=reward,
reasoning=action.memo or "",
next_state_summary=post_state,
))
# ββ STEP 11: Build observation and LOOP ββ
return self._build_obs(
reward=reward,
advisor_memos=advisor_memos,
advisor_messages=advisor_messages,
late_fee=late_fee_total,
interest=interest_total,
)
# βββββββββββββββββββββββββββββββββββββββββββββββββββ
# HELPER METHODS
# βββββββββββββββββββββββββββββββββββββββββββββββββββ
def _run_advisors(self):
"""Steps 3-4: Each agent observes partial state and produces a memo."""
world_hints = self.world_model.get_risk_hints(self.day)
# Run Revenue Agent first to get projection (now aware of market stress)
rev_memo = revenue_agent(
self.receivables,
self.invoices,
self.cash,
self.day,
market_stress=world_hints.get("market_stress", 0.0)
)
# Pass net 3-day position to Expenditure Agent for smarter prioritisation
exp_memo = expenditure_agent(
[inv for inv in self.invoices if inv.status != "paid"],
self.cash,
revenue_projection=rev_memo.get("net_3day_position", self.cash)
)
risk_memo = risk_agent(
self.cash,
exp_memo.get("total_outstanding", 0.0),
self.credit_used,
self.credit_limit,
world_hints
)
advisor_memos = {
"Expenditure": exp_memo,
"Revenue": rev_memo,
"Risk": risk_memo,
}
advisor_messages = {
"Expenditure": format_memo("Expenditure", exp_memo),
"Revenue": format_memo("Revenue", rev_memo),
"Risk": format_memo("Risk", risk_memo),
}
return advisor_memos, advisor_messages
def _find_invoice(self, invoice_id: str) -> Optional[Invoice]:
return next((i for i in self.invoices if i.id == invoice_id), None)
def _update_cash(self, amount: float):
self.cash += amount
if self.cash < 0:
self.credit_used += abs(self.cash)
self.cash = 0.0
def _check_day_end(self) -> bool:
# Day ends every 3 steps (morning, afternoon, evening review cycles)
return self._state.step_count % 3 == 0
def _build_obs(self, reward=0.0, advisor_memos=None, advisor_messages=None,
late_fee=0.0, interest=0.0) -> CashflowmanagerObservation:
active_invoices = [inv for inv in self.invoices if inv.status != "paid"]
vendor_info = {
vid: {"name": v.name, "trust_score": round(v.trust_score, 2)}
for vid, v in self.vendors.items()
}
return CashflowmanagerObservation(
day=self.day,
cash=round(self.cash, 2),
credit_used=round(self.credit_used, 2),
credit_limit=self.credit_limit,
invoices=active_invoices,
receivables=self.receivables,
vendor_profiles=vendor_info,
advisor_memos=advisor_memos or {},
advisor_messages=advisor_messages or {},
negotiation_result=self.last_negotiation,
world_events=self.last_world_events,
reward=round(reward, 4),
done=self.done,
metadata={
"step": self._state.step_count,
"late_fee": round(late_fee, 2),
"interest": round(interest, 4),
"day_ended": self._state.step_count > 0 and self._state.step_count % 3 == 0,
"world_events_count": len(self.world_model.get_triggered_events()),
},
)
def get_transitions(self) -> List[Dict]:
"""Return all stored transitions for training data export."""
return [t.model_dump() for t in self.transitions]
@property
def state(self) -> State:
return self._state |