feat: implement token efficiency optimization for issue #24
Implement comprehensive token efficiency system to reduce LLM costs: - Add prompt_optimizer.py: Token counting, compression, abbreviations - Add context_selector.py: Smart L1-L7 context layer selection - Add summarizer.py: Historical data aggregation and summarization - Add cache.py: TTL-based response caching with hit rate tracking - Enhance gemini_client.py: Integrate optimization, caching, metrics Key features: - Compressed prompts with abbreviations (40-50% reduction) - Smart context selection (L7 for normal, L6-L5 for strategic) - Response caching for HOLD decisions and high-confidence calls - Token usage tracking and metrics (avg tokens, cache hit rate) - Comprehensive test coverage (34 tests, 84-93% coverage) Metrics tracked: - Total tokens used - Avg tokens per decision - Cache hit rate - Cost per decision All tests passing (191 total, 76% overall coverage). Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
This commit is contained in:
agentson
293
src/brain/cache.py
Normal file
293
src/brain/cache.py
Normal file
@@ -0,0 +1,293 @@
|
||||
"""Response caching system for reducing redundant LLM calls.
|
||||
|
||||
This module provides caching for common trading scenarios:
|
||||
- TTL-based cache invalidation
|
||||
- Cache key based on market conditions
|
||||
- Cache hit rate monitoring
|
||||
- Special handling for HOLD decisions in quiet markets
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import hashlib
|
||||
import json
|
||||
import logging
|
||||
import time
|
||||
from dataclasses import dataclass, field
|
||||
from typing import Any, TYPE_CHECKING
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from src.brain.gemini_client import TradeDecision
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
@dataclass
|
||||
class CacheEntry:
|
||||
"""Cached decision with metadata."""
|
||||
|
||||
decision: "TradeDecision"
|
||||
cached_at: float # Unix timestamp
|
||||
hit_count: int = 0
|
||||
market_data_hash: str = ""
|
||||
|
||||
|
||||
@dataclass
|
||||
class CacheMetrics:
|
||||
"""Metrics for cache performance monitoring."""
|
||||
|
||||
total_requests: int = 0
|
||||
cache_hits: int = 0
|
||||
cache_misses: int = 0
|
||||
evictions: int = 0
|
||||
total_entries: int = 0
|
||||
|
||||
@property
|
||||
def hit_rate(self) -> float:
|
||||
"""Calculate cache hit rate."""
|
||||
if self.total_requests == 0:
|
||||
return 0.0
|
||||
return self.cache_hits / self.total_requests
|
||||
|
||||
def to_dict(self) -> dict[str, Any]:
|
||||
"""Convert metrics to dictionary."""
|
||||
return {
|
||||
"total_requests": self.total_requests,
|
||||
"cache_hits": self.cache_hits,
|
||||
"cache_misses": self.cache_misses,
|
||||
"hit_rate": self.hit_rate,
|
||||
"evictions": self.evictions,
|
||||
"total_entries": self.total_entries,
|
||||
}
|
||||
|
||||
|
||||
class DecisionCache:
|
||||
"""TTL-based cache for trade decisions."""
|
||||
|
||||
def __init__(self, ttl_seconds: int = 300, max_size: int = 1000) -> None:
|
||||
"""Initialize the decision cache.
|
||||
|
||||
Args:
|
||||
ttl_seconds: Time-to-live for cache entries in seconds (default: 5 minutes)
|
||||
max_size: Maximum number of cache entries
|
||||
"""
|
||||
self.ttl_seconds = ttl_seconds
|
||||
self.max_size = max_size
|
||||
self._cache: dict[str, CacheEntry] = {}
|
||||
self._metrics = CacheMetrics()
|
||||
|
||||
def _generate_cache_key(self, market_data: dict[str, Any]) -> str:
|
||||
"""Generate cache key from market data.
|
||||
|
||||
Key is based on:
|
||||
- Stock code
|
||||
- Current price (rounded to reduce sensitivity)
|
||||
- Market conditions (orderbook snapshot)
|
||||
|
||||
Args:
|
||||
market_data: Market data dictionary
|
||||
|
||||
Returns:
|
||||
Cache key string
|
||||
"""
|
||||
# Extract key components
|
||||
stock_code = market_data.get("stock_code", "UNKNOWN")
|
||||
current_price = market_data.get("current_price", 0)
|
||||
|
||||
# Round price to reduce sensitivity (cache hits for similar prices)
|
||||
# For prices > 1000, round to nearest 10
|
||||
# For prices < 1000, round to nearest 1
|
||||
if current_price > 1000:
|
||||
price_rounded = round(current_price / 10) * 10
|
||||
else:
|
||||
price_rounded = round(current_price)
|
||||
|
||||
# Include orderbook snapshot (if available)
|
||||
orderbook_key = ""
|
||||
if "orderbook" in market_data and market_data["orderbook"]:
|
||||
ob = market_data["orderbook"]
|
||||
# Just use bid/ask spread as indicator
|
||||
if "bid" in ob and "ask" in ob and ob["bid"] and ob["ask"]:
|
||||
bid_price = ob["bid"][0].get("price", 0) if ob["bid"] else 0
|
||||
ask_price = ob["ask"][0].get("price", 0) if ob["ask"] else 0
|
||||
spread = ask_price - bid_price
|
||||
orderbook_key = f"_spread{spread}"
|
||||
|
||||
# Generate cache key
|
||||
key_str = f"{stock_code}_{price_rounded}{orderbook_key}"
|
||||
|
||||
return key_str
|
||||
|
||||
def _generate_market_hash(self, market_data: dict[str, Any]) -> str:
|
||||
"""Generate hash of full market data for invalidation checks.
|
||||
|
||||
Args:
|
||||
market_data: Market data dictionary
|
||||
|
||||
Returns:
|
||||
Hash string
|
||||
"""
|
||||
# Create stable JSON representation
|
||||
stable_json = json.dumps(market_data, sort_keys=True, ensure_ascii=False)
|
||||
return hashlib.md5(stable_json.encode()).hexdigest()
|
||||
|
||||
def get(self, market_data: dict[str, Any]) -> TradeDecision | None:
|
||||
"""Retrieve cached decision if valid.
|
||||
|
||||
Args:
|
||||
market_data: Market data dictionary
|
||||
|
||||
Returns:
|
||||
Cached TradeDecision if valid, None otherwise
|
||||
"""
|
||||
self._metrics.total_requests += 1
|
||||
|
||||
cache_key = self._generate_cache_key(market_data)
|
||||
|
||||
if cache_key not in self._cache:
|
||||
self._metrics.cache_misses += 1
|
||||
return None
|
||||
|
||||
entry = self._cache[cache_key]
|
||||
current_time = time.time()
|
||||
|
||||
# Check TTL
|
||||
if current_time - entry.cached_at > self.ttl_seconds:
|
||||
# Expired
|
||||
del self._cache[cache_key]
|
||||
self._metrics.cache_misses += 1
|
||||
self._metrics.evictions += 1
|
||||
logger.debug("Cache expired for key: %s", cache_key)
|
||||
return None
|
||||
|
||||
# Cache hit
|
||||
entry.hit_count += 1
|
||||
self._metrics.cache_hits += 1
|
||||
logger.debug("Cache hit for key: %s (hits: %d)", cache_key, entry.hit_count)
|
||||
|
||||
return entry.decision
|
||||
|
||||
def set(
|
||||
self,
|
||||
market_data: dict[str, Any],
|
||||
decision: TradeDecision,
|
||||
) -> None:
|
||||
"""Store decision in cache.
|
||||
|
||||
Args:
|
||||
market_data: Market data dictionary
|
||||
decision: TradeDecision to cache
|
||||
"""
|
||||
cache_key = self._generate_cache_key(market_data)
|
||||
market_hash = self._generate_market_hash(market_data)
|
||||
|
||||
# Enforce max size (evict oldest if full)
|
||||
if len(self._cache) >= self.max_size:
|
||||
# Find oldest entry
|
||||
oldest_key = min(self._cache.keys(), key=lambda k: self._cache[k].cached_at)
|
||||
del self._cache[oldest_key]
|
||||
self._metrics.evictions += 1
|
||||
logger.debug("Cache full, evicted key: %s", oldest_key)
|
||||
|
||||
# Store entry
|
||||
entry = CacheEntry(
|
||||
decision=decision,
|
||||
cached_at=time.time(),
|
||||
market_data_hash=market_hash,
|
||||
)
|
||||
self._cache[cache_key] = entry
|
||||
self._metrics.total_entries = len(self._cache)
|
||||
|
||||
logger.debug("Cached decision for key: %s", cache_key)
|
||||
|
||||
def invalidate(self, stock_code: str | None = None) -> int:
|
||||
"""Invalidate cache entries.
|
||||
|
||||
Args:
|
||||
stock_code: Specific stock code to invalidate, or None for all
|
||||
|
||||
Returns:
|
||||
Number of entries invalidated
|
||||
"""
|
||||
if stock_code is None:
|
||||
# Clear all
|
||||
count = len(self._cache)
|
||||
self._cache.clear()
|
||||
self._metrics.evictions += count
|
||||
self._metrics.total_entries = 0
|
||||
logger.info("Invalidated all cache entries (%d)", count)
|
||||
return count
|
||||
|
||||
# Invalidate specific stock
|
||||
keys_to_remove = [k for k in self._cache.keys() if k.startswith(f"{stock_code}_")]
|
||||
count = len(keys_to_remove)
|
||||
|
||||
for key in keys_to_remove:
|
||||
del self._cache[key]
|
||||
|
||||
self._metrics.evictions += count
|
||||
self._metrics.total_entries = len(self._cache)
|
||||
logger.info("Invalidated %d cache entries for stock: %s", count, stock_code)
|
||||
|
||||
return count
|
||||
|
||||
def cleanup_expired(self) -> int:
|
||||
"""Remove expired entries from cache.
|
||||
|
||||
Returns:
|
||||
Number of entries removed
|
||||
"""
|
||||
current_time = time.time()
|
||||
expired_keys = [
|
||||
k
|
||||
for k, v in self._cache.items()
|
||||
if current_time - v.cached_at > self.ttl_seconds
|
||||
]
|
||||
|
||||
count = len(expired_keys)
|
||||
for key in expired_keys:
|
||||
del self._cache[key]
|
||||
|
||||
self._metrics.evictions += count
|
||||
self._metrics.total_entries = len(self._cache)
|
||||
|
||||
if count > 0:
|
||||
logger.debug("Cleaned up %d expired cache entries", count)
|
||||
|
||||
return count
|
||||
|
||||
def get_metrics(self) -> CacheMetrics:
|
||||
"""Get current cache metrics.
|
||||
|
||||
Returns:
|
||||
CacheMetrics object with current statistics
|
||||
"""
|
||||
return self._metrics
|
||||
|
||||
def reset_metrics(self) -> None:
|
||||
"""Reset cache metrics."""
|
||||
self._metrics = CacheMetrics(total_entries=len(self._cache))
|
||||
logger.info("Cache metrics reset")
|
||||
|
||||
def should_cache_decision(self, decision: TradeDecision) -> bool:
|
||||
"""Determine if a decision should be cached.
|
||||
|
||||
HOLD decisions with low confidence are good candidates for caching,
|
||||
as they're likely to recur in quiet markets.
|
||||
|
||||
Args:
|
||||
decision: TradeDecision to evaluate
|
||||
|
||||
Returns:
|
||||
True if decision should be cached
|
||||
"""
|
||||
# Cache HOLD decisions (common in quiet markets)
|
||||
if decision.action == "HOLD":
|
||||
return True
|
||||
|
||||
# Cache high-confidence decisions (stable signals)
|
||||
if decision.confidence >= 90:
|
||||
return True
|
||||
|
||||
# Don't cache low-confidence BUY/SELL (volatile signals)
|
||||
return False
|
||||
Reference in New Issue
Block a user