Initial TradeBot implementation
This commit is contained in:
Курнат Андрей
@@ -0,0 +1,226 @@
|
||||
from __future__ import annotations
|
||||
|
||||
import json
|
||||
from dataclasses import asdict, dataclass, field
|
||||
from typing import Any
|
||||
|
||||
import requests
|
||||
|
||||
from crypto_spot_bot.config import Settings
|
||||
from crypto_spot_bot.models import Candle, Ticker, utc_now
|
||||
from crypto_spot_bot.storage import Storage
|
||||
|
||||
|
||||
REGIMES = {"uptrend", "downtrend", "range", "breakout", "breakdown", "panic", "unknown"}
|
||||
RISK_LEVELS = {"low", "medium", "high"}
|
||||
|
||||
|
||||
@dataclass(slots=True)
|
||||
class LlmAdvice:
|
||||
symbol: str
|
||||
enabled: bool
|
||||
model: str
|
||||
market_regime: str = "unknown"
|
||||
risk_level: str = "medium"
|
||||
confidence_adjustment: float = 0.0
|
||||
block_entry: bool = False
|
||||
grid_suitable: bool = False
|
||||
reason_ru: str = "LLM Advisor не дал активной поправки."
|
||||
error: str = ""
|
||||
created_at: str = field(default_factory=lambda: utc_now().isoformat())
|
||||
|
||||
def as_dict(self) -> dict[str, Any]:
|
||||
return asdict(self)
|
||||
|
||||
|
||||
class LlmAdvisor:
|
||||
def __init__(self, settings: Settings, storage: Storage):
|
||||
self.settings = settings
|
||||
self.storage = storage
|
||||
self._cache: dict[str, LlmAdvice] = {}
|
||||
|
||||
def advice_for(
|
||||
self,
|
||||
*,
|
||||
symbol: str,
|
||||
candles: list[Candle],
|
||||
ticker: Ticker | None,
|
||||
pattern: dict[str, Any],
|
||||
learning: dict[str, Any],
|
||||
open_positions_for_symbol: int,
|
||||
account: dict[str, float],
|
||||
) -> LlmAdvice:
|
||||
if not self.settings.llm_advisor_enabled:
|
||||
return LlmAdvice(
|
||||
symbol=symbol,
|
||||
enabled=False,
|
||||
model=self.settings.ollama_model,
|
||||
reason_ru="LLM Advisor выключен.",
|
||||
)
|
||||
cached = self._cache.get(symbol)
|
||||
if cached and _age_seconds(cached.created_at) < self.settings.llm_advisor_min_interval_seconds:
|
||||
return cached
|
||||
|
||||
context = _build_context(symbol, candles, ticker, pattern, learning, open_positions_for_symbol, account)
|
||||
prompt = _prompt(context, self.settings.llm_advisor_max_adjustment)
|
||||
response_text = ""
|
||||
error = ""
|
||||
try:
|
||||
response = requests.post(
|
||||
f"{self.settings.ollama_base_url}/api/generate",
|
||||
json={
|
||||
"model": self.settings.ollama_model,
|
||||
"prompt": prompt,
|
||||
"stream": False,
|
||||
"options": {"temperature": 0.1},
|
||||
},
|
||||
timeout=self.settings.llm_advisor_timeout_seconds,
|
||||
)
|
||||
response.raise_for_status()
|
||||
payload = response.json()
|
||||
response_text = str(payload.get("response", ""))
|
||||
advice = self._parse(symbol, response_text)
|
||||
except Exception as exc:
|
||||
error = str(exc)
|
||||
advice = LlmAdvice(
|
||||
symbol=symbol,
|
||||
enabled=True,
|
||||
model=self.settings.ollama_model,
|
||||
reason_ru="LLM Advisor временно недоступен; используется нейтральная поправка.",
|
||||
error=error,
|
||||
)
|
||||
self._cache[symbol] = advice
|
||||
self.storage.insert_llm_advice(
|
||||
symbol=symbol,
|
||||
model=self.settings.ollama_model,
|
||||
prompt_json=context,
|
||||
response_text=response_text,
|
||||
advice_json=advice.as_dict(),
|
||||
error=error or advice.error,
|
||||
)
|
||||
return advice
|
||||
|
||||
def snapshot(self) -> dict[str, Any]:
|
||||
return {
|
||||
"enabled": self.settings.llm_advisor_enabled,
|
||||
"base_url": self.settings.ollama_base_url,
|
||||
"model": self.settings.ollama_model,
|
||||
"min_interval_seconds": self.settings.llm_advisor_min_interval_seconds,
|
||||
"max_adjustment": self.settings.llm_advisor_max_adjustment,
|
||||
"items": [advice.as_dict() for advice in self._cache.values()],
|
||||
}
|
||||
|
||||
def _parse(self, symbol: str, response_text: str) -> LlmAdvice:
|
||||
data = _extract_json(response_text)
|
||||
regime = str(data.get("market_regime", "unknown")).strip().lower()
|
||||
risk = str(data.get("risk_level", "medium")).strip().lower()
|
||||
adjustment = _clamp_float(
|
||||
data.get("confidence_adjustment", 0.0),
|
||||
-self.settings.llm_advisor_max_adjustment,
|
||||
self.settings.llm_advisor_max_adjustment,
|
||||
)
|
||||
return LlmAdvice(
|
||||
symbol=symbol,
|
||||
enabled=True,
|
||||
model=self.settings.ollama_model,
|
||||
market_regime=regime if regime in REGIMES else "unknown",
|
||||
risk_level=risk if risk in RISK_LEVELS else "medium",
|
||||
confidence_adjustment=adjustment,
|
||||
block_entry=bool(data.get("block_entry", False)),
|
||||
grid_suitable=bool(data.get("grid_suitable", False)),
|
||||
reason_ru=str(data.get("reason_ru", "LLM Advisor не объяснил вывод."))[:240],
|
||||
)
|
||||
|
||||
|
||||
def _build_context(
|
||||
symbol: str,
|
||||
candles: list[Candle],
|
||||
ticker: Ticker | None,
|
||||
pattern: dict[str, Any],
|
||||
learning: dict[str, Any],
|
||||
open_positions_for_symbol: int,
|
||||
account: dict[str, float],
|
||||
) -> dict[str, Any]:
|
||||
latest = candles[-1] if candles else None
|
||||
return {
|
||||
"mode": "paper_demo_only",
|
||||
"symbol": symbol,
|
||||
"objective": "reduce avoidable losing spot-long entries; do not promise profit",
|
||||
"market": {
|
||||
"last_price": ticker.last_price if ticker else None,
|
||||
"spread_percent": ticker.spread_percent if ticker else None,
|
||||
"turnover_24h": ticker.turnover_24h if ticker else None,
|
||||
"change_24h": ticker.change_24h if ticker else None,
|
||||
"close": latest.close if latest else None,
|
||||
"rsi_14": latest.rsi_14 if latest else None,
|
||||
"ema_20": latest.ema_20 if latest else None,
|
||||
"ema_50": latest.ema_50 if latest else None,
|
||||
"ema_200": latest.ema_200 if latest else None,
|
||||
"atr_14": latest.atr_14 if latest else None,
|
||||
"volume": latest.volume if latest else None,
|
||||
"volume_ma_20": latest.volume_ma_20 if latest else None,
|
||||
},
|
||||
"pattern": pattern,
|
||||
"learning": learning,
|
||||
"risk_state": {
|
||||
"equity": account.get("equity"),
|
||||
"cash": account.get("cash"),
|
||||
"exposure": account.get("exposure"),
|
||||
"drawdown": account.get("drawdown"),
|
||||
"open_positions_for_symbol": open_positions_for_symbol,
|
||||
},
|
||||
"allowed_output": {
|
||||
"market_regime": sorted(REGIMES),
|
||||
"risk_level": sorted(RISK_LEVELS),
|
||||
"confidence_adjustment": "number within configured bounds",
|
||||
"block_entry": "boolean; can only block buy, never force buy",
|
||||
"grid_suitable": "boolean",
|
||||
"reason_ru": "short Russian explanation",
|
||||
},
|
||||
}
|
||||
|
||||
|
||||
def _prompt(context: dict[str, Any], max_adjustment: float) -> str:
|
||||
return (
|
||||
"Ты LLM Advisor для paper-only crypto spot LONG бота. "
|
||||
"Ты не открываешь сделки и не обещаешь прибыль. "
|
||||
"Верни только валидный JSON без markdown. "
|
||||
f"confidence_adjustment должен быть от {-max_adjustment:.4f} до {max_adjustment:.4f}. "
|
||||
"Если рынок падающий, шаблон отрицательный или обучение убыточное, используй отрицательную поправку или block_entry=true. "
|
||||
"Если боковик и риск умеренный, можешь отметить grid_suitable=true. "
|
||||
"JSON keys: market_regime, risk_level, confidence_adjustment, block_entry, grid_suitable, reason_ru. "
|
||||
f"Context: {json.dumps(context, ensure_ascii=False, separators=(',', ':'))}"
|
||||
)
|
||||
|
||||
|
||||
def _extract_json(text: str) -> dict[str, Any]:
|
||||
stripped = text.strip()
|
||||
if stripped.startswith("```"):
|
||||
stripped = stripped.strip("`").strip()
|
||||
if stripped.lower().startswith("json"):
|
||||
stripped = stripped[4:].strip()
|
||||
try:
|
||||
data = json.loads(stripped)
|
||||
except json.JSONDecodeError:
|
||||
start = stripped.find("{")
|
||||
end = stripped.rfind("}")
|
||||
if start < 0 or end <= start:
|
||||
raise
|
||||
data = json.loads(stripped[start : end + 1])
|
||||
if not isinstance(data, dict):
|
||||
raise ValueError("LLM response JSON is not an object")
|
||||
return data
|
||||
|
||||
|
||||
def _clamp_float(value: Any, low: float, high: float) -> float:
|
||||
try:
|
||||
parsed = float(value)
|
||||
except (TypeError, ValueError):
|
||||
parsed = 0.0
|
||||
return round(max(low, min(high, parsed)), 4)
|
||||
|
||||
|
||||
def _age_seconds(created_at: str) -> float:
|
||||
from datetime import datetime
|
||||
|
||||
return (utc_now() - datetime.fromisoformat(created_at)).total_seconds()
|
||||
Reference in New Issue
Block a user