Spaces:
Runtime error
Runtime error
File size: 7,450 Bytes
f70ac6a | 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 | """
GraphAnalysisAgent - Network pattern detection
Analyzes account network: peers, cycles, communities
"""
from typing import Dict, List
from .base import Agent, AgentConfig, AgentResult
import time
class GraphAnalysisAgent(Agent):
"""
Analyzes transaction network around an account
Direct graph operations (no LLM needed for retrieval)
"""
def __init__(self, api_pool, app_state=None):
config = AgentConfig(
name="GraphAnalysisAgent",
model="llama-3.1-8b-instant", # llama-3.1-70b-versatile deprecated
temperature=0.2,
max_tokens=1000,
timeout_ms=10000,
)
super().__init__(config, api_pool)
self.app_state = app_state
def set_app_state(self, app_state):
self.app_state = app_state
def _build_prompt(self, **inputs) -> str:
return ""
def _parse_response(self, response_text: str) -> Dict:
return {}
async def invoke(self, account_id: str = None, hops: int = 2, max_nodes: int = 30, **kwargs) -> AgentResult:
"""
Analyze network around account
Returns: peers, cycles, community info, network metrics
"""
start_time = time.time()
if not self.app_state or not account_id:
return await self._create_result(
success=False,
data={},
error="No account_id or app_state",
start_time=start_time,
)
try:
data = {
"account_id": account_id,
"network_metrics": {},
"peers": [],
"cycles": [],
"community": {},
"subgraph_summary": {},
}
if not hasattr(self.app_state, "graph") or self.app_state.graph is None:
return await self._create_result(
success=False,
data=data,
error="Graph not available",
start_time=start_time,
)
graph = self.app_state.graph
if account_id not in graph.nodes():
return await self._create_result(
success=True,
data=data,
start_time=start_time,
)
# Get network metrics
data["network_metrics"] = {
"in_degree": graph.in_degree(account_id),
"out_degree": graph.out_degree(account_id),
"total_degree": graph.degree(account_id),
}
# Add centrality scores if available
if hasattr(self.app_state, "pagerank_scores"):
data["network_metrics"]["pagerank"] = self.app_state.pagerank_scores.get(account_id, 0)
if hasattr(self.app_state, "betweenness_scores"):
data["network_metrics"]["betweenness"] = self.app_state.betweenness_scores.get(account_id, 0)
# Community info
if hasattr(self.app_state, "louvain_partition"):
community_id = self.app_state.louvain_partition.get(account_id)
if community_id is not None:
community_members = [
node for node, comm in self.app_state.louvain_partition.items()
if comm == community_id
]
fraud_in_community = sum(
1 for node in community_members
if self.app_state.features_by_account.get(node, {}).get("fraud_flag", 0)
)
data["community"] = {
"community_id": community_id,
"size": len(community_members),
"fraud_count": fraud_in_community,
"fraud_rate": fraud_in_community / max(1, len(community_members)),
}
# Get direct peers (1-hop)
successors = list(graph.successors(account_id))[:10]
predecessors = list(graph.predecessors(account_id))[:10]
peers = []
for peer_id in set(successors + predecessors):
peer_features = self.app_state.features_by_account.get(peer_id, {})
peer_data = {
"account_id": peer_id,
"relationship": [],
"risk_score": peer_features.get("risk_score", 0),
"is_fraud": peer_features.get("fraud_flag", 0) == 1,
}
if peer_id in successors:
peer_data["relationship"].append("sends_to")
if peer_id in predecessors:
peer_data["relationship"].append("receives_from")
peers.append(peer_data)
# Sort peers by risk
peers.sort(key=lambda x: x["risk_score"], reverse=True)
data["peers"] = peers[:10]
# Detect cycles (simple cycle detection)
cycles = self._detect_cycles(graph, account_id, max_length=5)
data["cycles"] = cycles[:5]
# Subgraph summary
try:
from src.graph_builder import get_subgraph
subgraph = get_subgraph(graph, account_id, hops=hops, max_nodes=max_nodes)
data["subgraph_summary"] = {
"nodes_count": subgraph.number_of_nodes(),
"edges_count": subgraph.number_of_edges(),
"depth": hops,
}
except Exception as e:
self.logger.warning(f"Subgraph extraction failed: {e}")
tokens_estimate = self._estimate_tokens(str(data))
self.logger.info(
f"[OK] {self.config.name}: Analyzed {account_id} "
f"({len(peers)} peers, {len(cycles)} cycles, ~{tokens_estimate} tokens)"
)
return await self._create_result(
success=True,
data=data,
tokens_output=tokens_estimate,
start_time=start_time,
)
except Exception as e:
self.logger.error(f"[FAIL] {self.config.name}: {e}")
return await self._create_result(
success=False,
data={},
error=str(e),
start_time=start_time,
)
def _detect_cycles(self, graph, account_id: str, max_length: int = 5) -> List[Dict]:
"""Detect cycles involving the account (simplified)"""
cycles = []
try:
import networkx as nx
# Get small subgraph for cycle detection (performance)
try:
from src.graph_builder import get_subgraph
subgraph = get_subgraph(graph, account_id, hops=3, max_nodes=50)
except Exception:
subgraph = graph
# Find simple cycles containing account_id
simple_cycles = list(nx.simple_cycles(subgraph))
for cycle in simple_cycles[:5]:
if account_id in cycle and len(cycle) <= max_length:
cycles.append({
"cycle": cycle + [cycle[0]], # Close the cycle for display
"length": len(cycle),
})
except Exception as e:
self.logger.warning(f"Cycle detection failed: {e}")
return cycles
|