Spaces:
Running
Running
File size: 6,955 Bytes
8ff817c |
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 |
"""
Router Agent: Decides how to improve an insufficient plan.
When the verifier determines the plan is insufficient, the router decides:
- "Add Step": Add a new step to the plan
- "Step N": Backtrack to step N and fix it
"""
import re
from langchain_core.messages import AIMessage
from ..utils.formatters import format_data_descriptions, format_plan, gemini_text
from ..utils.state import DSStarState
def router_node(state: DSStarState) -> dict:
"""
Router Agent Node: Decides how to improve the plan.
Analyzes the current situation and determines whether to:
1. Add a new step to the plan
2. Backtrack and fix an existing step
Args:
state: Current DSStarState
Returns:
Dictionary with updated state fields:
- router_decision: "Add Step" or "Step N"
- iteration: Incremented iteration count
- messages: Agent communication messages
- next: "planner" (add step) or "backtrack" (fix step)
"""
print("=" * 60)
print("ROUTER AGENT STARTING...")
print("=" * 60)
data_context = format_data_descriptions(state["data_descriptions"])
plan_text = format_plan(state["plan"])
prompt = f"""You are an expert data analyst router.
The current plan is INSUFFICIENT to answer the question.
Original Question: {state["query"]}
Available Data:
{data_context}
Current Plan:
{plan_text}
Execution Result:
{state["execution_result"][:500]}
Task: Decide how to improve the plan:
1. If a current step is WRONG or needs fixing: Answer "Step N" (where N is the step number, e.g., "Step 2")
2. If we need to ADD a NEW step: Answer "Add Step"
Answer with ONLY: "Step 1", "Step 2", etc. OR "Add Step"
No explanation needed."""
try:
# Get LLM response
response = state["llm"].invoke(prompt)
# Handle different response formats
if hasattr(response, "content") and isinstance(response.content, list):
response_text = gemini_text(response)
elif hasattr(response, "content"):
response_text = response.content
else:
response_text = str(response)
# Parse decision
response_lower = response_text.strip().lower()
if "add step" in response_lower:
decision = "Add Step"
next_node = "planner"
else:
# Try to extract step number
match = re.search(r"step\s+(\d+)", response_lower)
if match:
decision = f"Step {match.group(1)}"
next_node = "backtrack"
else:
# Default to adding new step
decision = "Add Step"
next_node = "planner"
print(f"\nRouter Decision: {decision}")
print(
f"Next Action: {'Backtrack' if next_node == 'backtrack' else 'Add New Step'}"
)
print("=" * 60)
return {
"router_decision": decision,
"messages": [AIMessage(content=f"Router: {decision}")],
"iteration": state["iteration"] + 1,
"next": next_node,
}
except Exception as e:
# On error, default to adding new step
print(f"\n✗ Router error: {str(e)}")
print("Defaulting to 'Add Step'")
return {
"router_decision": "Add Step",
"messages": [AIMessage(content=f"Router error, adding step: {str(e)}")],
"iteration": state["iteration"] + 1,
"next": "planner",
}
def backtrack_node(state: DSStarState) -> dict:
"""
Backtrack Node: Truncates plan to remove incorrect steps.
When router identifies a wrong step, this node:
1. Parses the step number from router_decision
2. Truncates the plan to remove that step and all subsequent steps
3. Routes back to planner to regenerate from that point
Args:
state: Current DSStarState
Returns:
Dictionary with updated state fields:
- plan: Truncated plan
- messages: Agent communication messages
- next: "planner" to regenerate from truncation point
"""
print("=" * 60)
print("BACKTRACK NODE ACTIVATING...")
print("=" * 60)
try:
# Extract step number from router decision
match = re.search(r"step\s+(\d+)", state["router_decision"].lower())
if match:
step_num = int(match.group(1))
else:
# If parsing fails, just add new step
print("Failed to parse step number, adding new step instead")
return {
"messages": [
AIMessage(content="Backtrack parsing failed, adding new step")
],
"next": "planner",
}
# Truncate plan to steps before the wrong one
# Keep steps 0 to (step_num - 2), which are steps 1 to (step_num - 1) in human counting
truncated_plan = state["plan"][: step_num - 1] if step_num > 1 else []
print(
f"Truncating plan from {len(state['plan'])} to {len(truncated_plan)} steps"
)
print(f"Removed step {step_num} and beyond")
print("=" * 60)
# Return the truncated plan (replaces entire plan, not appends)
return {
"plan": truncated_plan,
"messages": [AIMessage(content=f"Backtracked to step {step_num - 1}")],
"next": "planner",
}
except Exception as e:
print(f"✗ Backtrack error: {str(e)}")
return {
"messages": [AIMessage(content=f"Backtrack error: {str(e)}, continuing")],
"next": "planner",
}
# Standalone test function
def test_router(
llm, query: str, data_descriptions: dict, plan: list, execution_result: str
):
"""
Test the router agent independently.
Args:
llm: LLM instance
query: User query
data_descriptions: Dict of filename -> description
plan: Current plan steps
execution_result: Result from code execution
Returns:
Dictionary with router results
"""
# Create minimal test state
test_state = {
"llm": llm,
"query": query,
"data_descriptions": data_descriptions,
"plan": plan,
"current_code": "",
"execution_result": execution_result,
"is_sufficient": False,
"router_decision": "",
"iteration": 0,
"max_iterations": 20,
"messages": [],
"next": "router",
}
result = router_node(test_state)
print("\n" + "=" * 60)
print("ROUTER TEST RESULTS")
print("=" * 60)
print(f"Decision: {result.get('router_decision', 'unknown')}")
print(f"Next Node: {result.get('next', 'unknown')}")
return result
|