Update context_parser.py
Browse files- context_parser.py +104 -33
context_parser.py
CHANGED
|
@@ -1,31 +1,115 @@
|
|
| 1 |
from __future__ import annotations
|
| 2 |
|
| 3 |
import re
|
| 4 |
-
from typing import Any, Dict, List, Optional
|
| 5 |
|
| 6 |
|
| 7 |
-
def
|
| 8 |
-
"""
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
|
|
|
| 12 |
|
| 13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
"""
|
| 15 |
raw = (text or "").strip()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
if not raw:
|
| 17 |
-
return
|
| 18 |
|
| 19 |
-
#
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
return
|
| 27 |
|
| 28 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 29 |
|
| 30 |
|
| 31 |
def _extract_options(text: str) -> List[str]:
|
|
@@ -42,7 +126,6 @@ def _extract_options(text: str) -> List[str]:
|
|
| 42 |
if options:
|
| 43 |
return options
|
| 44 |
|
| 45 |
-
# fallback: inline A) ... B) ...
|
| 46 |
matches = re.findall(r"(?:^|\s)([A-E])[\)\.\:]\s*(.*?)(?=(?:\s+[A-E][\)\.\:])|$)", text, flags=re.I | re.S)
|
| 47 |
if matches:
|
| 48 |
return [m[1].strip() for m in matches if m[1].strip()]
|
|
@@ -51,10 +134,6 @@ def _extract_options(text: str) -> List[str]:
|
|
| 51 |
|
| 52 |
|
| 53 |
def extract_game_context_fields(text: str) -> Dict[str, Any]:
|
| 54 |
-
"""
|
| 55 |
-
Extracts lightweight structured fields from hidden Unity/game context.
|
| 56 |
-
Always returns stable keys expected by app.py.
|
| 57 |
-
"""
|
| 58 |
raw = (text or "").strip()
|
| 59 |
|
| 60 |
result: Dict[str, Any] = {
|
|
@@ -70,33 +149,27 @@ def extract_game_context_fields(text: str) -> Dict[str, Any]:
|
|
| 70 |
if not raw:
|
| 71 |
return result
|
| 72 |
|
| 73 |
-
# question
|
| 74 |
q_match = re.search(r"\bquestion\s*[:=]\s*(.+?)(?=\n[A-Za-z_ ]+\s*[:=]|\Z)", raw, flags=re.I | re.S)
|
| 75 |
if q_match:
|
| 76 |
result["question"] = q_match.group(1).strip()
|
| 77 |
|
| 78 |
-
# options block
|
| 79 |
opt_match = re.search(r"\b(?:options|choices|answers)\s*[:=]\s*(.+?)(?=\n[A-Za-z_ ]+\s*[:=]|\Z)", raw, flags=re.I | re.S)
|
| 80 |
if opt_match:
|
| 81 |
result["options"] = _extract_options(opt_match.group(1))
|
| 82 |
|
| 83 |
-
# if no explicit options block, scan whole context
|
| 84 |
if not result["options"]:
|
| 85 |
result["options"] = _extract_options(raw)
|
| 86 |
|
| 87 |
result["has_choices"] = len(result["options"]) > 0
|
| 88 |
|
| 89 |
-
# difficulty
|
| 90 |
difficulty_match = re.search(r"\bdifficulty\s*[:=]\s*([A-Za-z0-9_\- ]+)", raw, flags=re.I)
|
| 91 |
if difficulty_match:
|
| 92 |
result["difficulty"] = difficulty_match.group(1).strip()
|
| 93 |
|
| 94 |
-
# category/topic
|
| 95 |
category_match = re.search(r"\b(?:category|topic)\s*[:=]\s*([A-Za-z0-9_\- /]+)", raw, flags=re.I)
|
| 96 |
if category_match:
|
| 97 |
result["category"] = category_match.group(1).strip()
|
| 98 |
|
| 99 |
-
# money/balance
|
| 100 |
money_match = re.search(r"\b(?:money|balance|bank)\s*[:=]\s*([\-]?\d+(?:\.\d+)?)", raw, flags=re.I)
|
| 101 |
if money_match:
|
| 102 |
try:
|
|
@@ -127,10 +200,6 @@ def extract_game_context_fields(text: str) -> Dict[str, Any]:
|
|
| 127 |
|
| 128 |
|
| 129 |
def detect_intent(text: str, incoming_help_mode: Optional[str] = None) -> str:
|
| 130 |
-
"""
|
| 131 |
-
Returns one of:
|
| 132 |
-
answer, hint, instruction, walkthrough, explain, method, definition, concept
|
| 133 |
-
"""
|
| 134 |
forced = (incoming_help_mode or "").strip().lower()
|
| 135 |
if forced in {
|
| 136 |
"answer",
|
|
@@ -213,8 +282,10 @@ def detect_intent(text: str, incoming_help_mode: Optional[str] = None) -> str:
|
|
| 213 |
|
| 214 |
|
| 215 |
def intent_to_help_mode(intent: str) -> str:
|
| 216 |
-
if intent in {"walkthrough", "step_by_step"
|
| 217 |
return "walkthrough"
|
|
|
|
|
|
|
| 218 |
if intent == "hint":
|
| 219 |
return "hint"
|
| 220 |
if intent in {"definition", "instruction"}:
|
|
|
|
| 1 |
from __future__ import annotations
|
| 2 |
|
| 3 |
import re
|
| 4 |
+
from typing import Any, Dict, List, Optional
|
| 5 |
|
| 6 |
|
| 7 |
+
def _to_bool(value: str) -> bool:
|
| 8 |
+
return str(value or "").strip().lower() in {"true", "1", "yes", "y"}
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
def _looks_like_int(value: str) -> bool:
|
| 12 |
+
return bool(re.fullmatch(r"-?\d+", str(value or "").strip()))
|
| 13 |
|
| 14 |
+
|
| 15 |
+
def split_unity_message(text: str) -> Dict[str, Any]:
|
| 16 |
+
"""
|
| 17 |
+
Parses several Unity-style message formats and always returns a stable dict.
|
| 18 |
+
|
| 19 |
+
Supported cases:
|
| 20 |
+
1. Plain user text
|
| 21 |
+
2. Hidden-context prefix + USER:/PROMPT:/MESSAGE:
|
| 22 |
+
3. Structured multiline payload like:
|
| 23 |
+
hint
|
| 24 |
+
x/5 = 12
|
| 25 |
+
0
|
| 26 |
+
solve
|
| 27 |
+
False
|
| 28 |
+
answer
|
| 29 |
+
answer
|
| 30 |
+
algebra
|
| 31 |
+
Quantitative
|
| 32 |
"""
|
| 33 |
raw = (text or "").strip()
|
| 34 |
+
|
| 35 |
+
result: Dict[str, Any] = {
|
| 36 |
+
"hidden_context": "",
|
| 37 |
+
"user_text": raw,
|
| 38 |
+
"question_text": "",
|
| 39 |
+
"hint_stage": 0,
|
| 40 |
+
"user_last_input_type": "",
|
| 41 |
+
"built_on_previous_turn": False,
|
| 42 |
+
"help_mode": "",
|
| 43 |
+
"intent": "",
|
| 44 |
+
"topic": "",
|
| 45 |
+
"category": "",
|
| 46 |
+
}
|
| 47 |
+
|
| 48 |
if not raw:
|
| 49 |
+
return result
|
| 50 |
|
| 51 |
+
# Case 1: hidden/system context followed by USER:/PROMPT:/MESSAGE:
|
| 52 |
+
tagged_match = re.search(r"(?is)^(.*?)(?:\buser\b|\bprompt\b|\bmessage\b)\s*:\s*(.+)$", raw)
|
| 53 |
+
if tagged_match:
|
| 54 |
+
hidden = (tagged_match.group(1) or "").strip()
|
| 55 |
+
user = (tagged_match.group(2) or "").strip()
|
| 56 |
+
result["hidden_context"] = hidden
|
| 57 |
+
result["user_text"] = user
|
| 58 |
+
return result
|
| 59 |
|
| 60 |
+
# Case 2: exact structured Unity payload block
|
| 61 |
+
lines = [line.strip() for line in raw.splitlines() if line.strip()]
|
| 62 |
+
if len(lines) >= 9 and _looks_like_int(lines[2]) and lines[4].lower() in {"true", "false"}:
|
| 63 |
+
result["user_text"] = lines[0]
|
| 64 |
+
result["question_text"] = lines[1]
|
| 65 |
+
result["hint_stage"] = int(lines[2])
|
| 66 |
+
result["user_last_input_type"] = lines[3]
|
| 67 |
+
result["built_on_previous_turn"] = _to_bool(lines[4])
|
| 68 |
+
result["help_mode"] = lines[5]
|
| 69 |
+
result["intent"] = lines[6]
|
| 70 |
+
result["topic"] = lines[7]
|
| 71 |
+
result["category"] = lines[8]
|
| 72 |
+
return result
|
| 73 |
+
|
| 74 |
+
# Case 3: field-based payload
|
| 75 |
+
def _extract_field(name: str) -> str:
|
| 76 |
+
match = re.search(rf"(?im)^\s*{re.escape(name)}\s*[:=]\s*(.+?)\s*$", raw)
|
| 77 |
+
return (match.group(1) or "").strip() if match else ""
|
| 78 |
+
|
| 79 |
+
question_text = _extract_field("question") or _extract_field("question_text")
|
| 80 |
+
user_text = _extract_field("user") or _extract_field("message") or _extract_field("prompt")
|
| 81 |
+
hint_stage_text = _extract_field("hint_stage")
|
| 82 |
+
user_last_input_type = _extract_field("user_last_input_type")
|
| 83 |
+
built_on_previous_turn = _extract_field("built_on_previous_turn")
|
| 84 |
+
help_mode = _extract_field("help_mode")
|
| 85 |
+
intent = _extract_field("intent")
|
| 86 |
+
topic = _extract_field("topic")
|
| 87 |
+
category = _extract_field("category")
|
| 88 |
+
|
| 89 |
+
if any([
|
| 90 |
+
question_text,
|
| 91 |
+
user_text,
|
| 92 |
+
hint_stage_text,
|
| 93 |
+
user_last_input_type,
|
| 94 |
+
built_on_previous_turn,
|
| 95 |
+
help_mode,
|
| 96 |
+
intent,
|
| 97 |
+
topic,
|
| 98 |
+
category,
|
| 99 |
+
]):
|
| 100 |
+
result["question_text"] = question_text
|
| 101 |
+
result["user_text"] = user_text or raw
|
| 102 |
+
result["hint_stage"] = int(hint_stage_text) if _looks_like_int(hint_stage_text) else 0
|
| 103 |
+
result["user_last_input_type"] = user_last_input_type
|
| 104 |
+
result["built_on_previous_turn"] = _to_bool(built_on_previous_turn)
|
| 105 |
+
result["help_mode"] = help_mode
|
| 106 |
+
result["intent"] = intent
|
| 107 |
+
result["topic"] = topic
|
| 108 |
+
result["category"] = category
|
| 109 |
+
return result
|
| 110 |
+
|
| 111 |
+
# Fallback: plain message
|
| 112 |
+
return result
|
| 113 |
|
| 114 |
|
| 115 |
def _extract_options(text: str) -> List[str]:
|
|
|
|
| 126 |
if options:
|
| 127 |
return options
|
| 128 |
|
|
|
|
| 129 |
matches = re.findall(r"(?:^|\s)([A-E])[\)\.\:]\s*(.*?)(?=(?:\s+[A-E][\)\.\:])|$)", text, flags=re.I | re.S)
|
| 130 |
if matches:
|
| 131 |
return [m[1].strip() for m in matches if m[1].strip()]
|
|
|
|
| 134 |
|
| 135 |
|
| 136 |
def extract_game_context_fields(text: str) -> Dict[str, Any]:
|
|
|
|
|
|
|
|
|
|
|
|
|
| 137 |
raw = (text or "").strip()
|
| 138 |
|
| 139 |
result: Dict[str, Any] = {
|
|
|
|
| 149 |
if not raw:
|
| 150 |
return result
|
| 151 |
|
|
|
|
| 152 |
q_match = re.search(r"\bquestion\s*[:=]\s*(.+?)(?=\n[A-Za-z_ ]+\s*[:=]|\Z)", raw, flags=re.I | re.S)
|
| 153 |
if q_match:
|
| 154 |
result["question"] = q_match.group(1).strip()
|
| 155 |
|
|
|
|
| 156 |
opt_match = re.search(r"\b(?:options|choices|answers)\s*[:=]\s*(.+?)(?=\n[A-Za-z_ ]+\s*[:=]|\Z)", raw, flags=re.I | re.S)
|
| 157 |
if opt_match:
|
| 158 |
result["options"] = _extract_options(opt_match.group(1))
|
| 159 |
|
|
|
|
| 160 |
if not result["options"]:
|
| 161 |
result["options"] = _extract_options(raw)
|
| 162 |
|
| 163 |
result["has_choices"] = len(result["options"]) > 0
|
| 164 |
|
|
|
|
| 165 |
difficulty_match = re.search(r"\bdifficulty\s*[:=]\s*([A-Za-z0-9_\- ]+)", raw, flags=re.I)
|
| 166 |
if difficulty_match:
|
| 167 |
result["difficulty"] = difficulty_match.group(1).strip()
|
| 168 |
|
|
|
|
| 169 |
category_match = re.search(r"\b(?:category|topic)\s*[:=]\s*([A-Za-z0-9_\- /]+)", raw, flags=re.I)
|
| 170 |
if category_match:
|
| 171 |
result["category"] = category_match.group(1).strip()
|
| 172 |
|
|
|
|
| 173 |
money_match = re.search(r"\b(?:money|balance|bank)\s*[:=]\s*([\-]?\d+(?:\.\d+)?)", raw, flags=re.I)
|
| 174 |
if money_match:
|
| 175 |
try:
|
|
|
|
| 200 |
|
| 201 |
|
| 202 |
def detect_intent(text: str, incoming_help_mode: Optional[str] = None) -> str:
|
|
|
|
|
|
|
|
|
|
|
|
|
| 203 |
forced = (incoming_help_mode or "").strip().lower()
|
| 204 |
if forced in {
|
| 205 |
"answer",
|
|
|
|
| 282 |
|
| 283 |
|
| 284 |
def intent_to_help_mode(intent: str) -> str:
|
| 285 |
+
if intent in {"walkthrough", "step_by_step"}:
|
| 286 |
return "walkthrough"
|
| 287 |
+
if intent in {"explain", "method", "concept"}:
|
| 288 |
+
return "explain"
|
| 289 |
if intent == "hint":
|
| 290 |
return "hint"
|
| 291 |
if intent in {"definition", "instruction"}:
|