WP13b Refactoring ingestion und Chunker
This commit is contained in:
1 parent
cf302e8334
commit
94e5ebf577
13 files changed
+607
-762
No files matched your search
Whitespace-only changes.
@@ -0,0 +1,31 @@
|
||||
"""
|
||||
FILE: app/core/chunking/chunking_models.py
|
||||
DESCRIPTION: Datenklassen für das Chunking-System.
|
||||
"""
|
||||
from dataclasses import dataclass, field
|
||||
from typing import List, Dict, Optional, Any
|
||||
|
||||
@dataclass
|
||||
class RawBlock:
|
||||
"""Repräsentiert einen logischen Block aus dem Markdown-Parsing."""
|
||||
kind: str
|
||||
text: str
|
||||
level: Optional[int]
|
||||
section_path: str
|
||||
section_title: Optional[str]
|
||||
|
||||
@dataclass
|
||||
class Chunk:
|
||||
"""Das finale Chunk-Objekt für Embedding und Graph-Speicherung."""
|
||||
id: str
|
||||
note_id: str
|
||||
index: int
|
||||
text: str
|
||||
window: str
|
||||
token_count: int
|
||||
section_title: Optional[str]
|
||||
section_path: str
|
||||
neighbors_prev: Optional[str]
|
||||
neighbors_next: Optional[str]
|
||||
candidate_pool: List[Dict[str, Any]] = field(default_factory=list)
|
||||
suggested_edges: Optional[List[str]] = None
|
||||
@@ -0,0 +1,74 @@
|
||||
"""
|
||||
FILE: app/core/chunking/chunking_parser.py
|
||||
DESCRIPTION: Zerlegt Markdown in Blöcke und extrahiert Kanten-Strings.
|
||||
"""
|
||||
import re
|
||||
from typing import List, Tuple, Set
|
||||
from .chunking_models import RawBlock
|
||||
from .chunking_utils import extract_frontmatter_from_text
|
||||
|
||||
_WS = re.compile(r'\s+')
|
||||
_SENT_SPLIT = re.compile(r'(?<=[.!?])\s+(?=[A-ZÄÖÜ0-9„(])')
|
||||
|
||||
def split_sentences(text: str) -> list[str]:
|
||||
"""Teilt Text in Sätze auf."""
|
||||
text = _WS.sub(' ', text.strip())
|
||||
if not text: return []
|
||||
return [p.strip() for p in _SENT_SPLIT.split(text) if p.strip()]
|
||||
|
||||
def parse_blocks(md_text: str) -> Tuple[List[RawBlock], str]:
|
||||
"""Zerlegt Text in logische Einheiten."""
|
||||
blocks = []
|
||||
h1_title = "Dokument"; section_path = "/"; current_h2 = None
|
||||
fm, text_without_fm = extract_frontmatter_from_text(md_text)
|
||||
h1_match = re.search(r'^#\s+(.*)', text_without_fm, re.MULTILINE)
|
||||
if h1_match: h1_title = h1_match.group(1).strip()
|
||||
lines = text_without_fm.split('\n')
|
||||
buffer = []
|
||||
for line in lines:
|
||||
stripped = line.strip()
|
||||
if stripped.startswith('# '): continue
|
||||
elif stripped.startswith('## '):
|
||||
if buffer:
|
||||
content = "\n".join(buffer).strip()
|
||||
if content: blocks.append(RawBlock("paragraph", content, None, section_path, current_h2))
|
||||
buffer = []
|
||||
current_h2 = stripped[3:].strip()
|
||||
section_path = f"/{current_h2}"
|
||||
blocks.append(RawBlock("heading", stripped, 2, section_path, current_h2))
|
||||
elif not stripped:
|
||||
if buffer:
|
||||
content = "\n".join(buffer).strip()
|
||||
if content: blocks.append(RawBlock("paragraph", content, None, section_path, current_h2))
|
||||
buffer = []
|
||||
else: buffer.append(line)
|
||||
if buffer:
|
||||
content = "\n".join(buffer).strip()
|
||||
if content: blocks.append(RawBlock("paragraph", content, None, section_path, current_h2))
|
||||
return blocks, h1_title
|
||||
|
||||
def parse_edges_robust(text: str) -> Set[str]:
|
||||
"""Extrahiert Kanten-Kandidaten (Wikilinks, Callouts)."""
|
||||
found_edges = set()
|
||||
inlines = re.findall(r'\[\[rel:([^\|\]]+)\|?([^\]]*)\]\]', text)
|
||||
for kind, target in inlines:
|
||||
k = kind.strip().lower()
|
||||
t = target.strip()
|
||||
if k and t: found_edges.add(f"{k}:{t}")
|
||||
lines = text.split('\n')
|
||||
current_edge_type = None
|
||||
for line in lines:
|
||||
stripped = line.strip()
|
||||
callout_match = re.match(r'>\s*\[!edge\]\s*([^:\s]+)', stripped)
|
||||
if callout_match:
|
||||
current_edge_type = callout_match.group(1).strip().lower()
|
||||
links = re.findall(r'\[\[([^\]]+)\]\]', stripped)
|
||||
for l in links:
|
||||
if "rel:" not in l: found_edges.add(f"{current_edge_type}:{l}")
|
||||
continue
|
||||
if current_edge_type and stripped.startswith('>'):
|
||||
links = re.findall(r'\[\[([^\]]+)\]\]', stripped)
|
||||
for l in links:
|
||||
if "rel:" not in l: found_edges.add(f"{current_edge_type}:{l}")
|
||||
elif not stripped.startswith('>'): current_edge_type = None
|
||||
return found_edges
|
||||
@@ -0,0 +1,25 @@
|
||||
"""
|
||||
FILE: app/core/chunking/chunking_propagation.py
|
||||
DESCRIPTION: Vererbung von Kanten (Inheritance) über Sektions-Pfade.
|
||||
"""
|
||||
from typing import List, Dict, Set
|
||||
from .chunking_models import Chunk, RawBlock
|
||||
from .chunking_parser import parse_edges_robust
|
||||
|
||||
def propagate_section_edges(chunks: List[Chunk], blocks: List[RawBlock]) -> List[Chunk]:
|
||||
"""WP-15b: Kanten aus Headings werden an Sub-Chunks vererbt."""
|
||||
section_inheritance: Dict[str, Set[str]] = {}
|
||||
for b in blocks:
|
||||
if b.kind == "heading":
|
||||
edges = parse_edges_robust(b.text)
|
||||
if edges:
|
||||
if b.section_path not in section_inheritance:
|
||||
section_inheritance[b.section_path] = set()
|
||||
section_inheritance[b.section_path].update(edges)
|
||||
|
||||
for ch in chunks:
|
||||
inherited = section_inheritance.get(ch.section_path, set())
|
||||
for e_str in inherited:
|
||||
kind, target = e_str.split(':', 1)
|
||||
ch.candidate_pool.append({"kind": kind, "to": target, "provenance": "inherited"})
|
||||
return chunks
|
||||
@@ -0,0 +1,74 @@
|
||||
"""
|
||||
FILE: app/core/chunking/chunking_strategies.py
|
||||
DESCRIPTION: Implementierung der mathematischen Splitting-Strategien.
|
||||
"""
|
||||
from typing import List, Dict, Any
|
||||
from .chunking_models import RawBlock, Chunk
|
||||
from .chunking_utils import estimate_tokens
|
||||
from .chunking_parser import split_sentences
|
||||
|
||||
def strategy_sliding_window(blocks: List[RawBlock], config: Dict[str, Any], note_id: str, context_prefix: str = "") -> List[Chunk]:
|
||||
"""Fasst Blöcke zusammen und schneidet bei 'target' Tokens."""
|
||||
target = config.get("target", 400); max_tokens = config.get("max", 600)
|
||||
overlap_val = config.get("overlap", (50, 80))
|
||||
overlap = sum(overlap_val) // 2 if isinstance(overlap_val, tuple) else overlap_val
|
||||
chunks = []; buf = []
|
||||
|
||||
def _add(txt, sec, path):
|
||||
idx = len(chunks); win = f"{context_prefix}\n{txt}".strip() if context_prefix else txt
|
||||
chunks.append(Chunk(id=f"{note_id}#c{idx:02d}", note_id=note_id, index=idx, text=txt, window=win, token_count=estimate_tokens(txt), section_title=sec, section_path=path, neighbors_prev=None, neighbors_next=None))
|
||||
|
||||
def flush():
|
||||
nonlocal buf
|
||||
if not buf: return
|
||||
text_body = "\n\n".join([b.text for b in buf])
|
||||
sec_title = buf[-1].section_title; sec_path = buf[-1].section_path
|
||||
if estimate_tokens(text_body) <= max_tokens: _add(text_body, sec_title, sec_path)
|
||||
else:
|
||||
sents = split_sentences(text_body); cur_sents = []; cur_len = 0
|
||||
for s in sents:
|
||||
slen = estimate_tokens(s)
|
||||
if cur_len + slen > target and cur_sents:
|
||||
_add(" ".join(cur_sents), sec_title, sec_path)
|
||||
ov_s = []; ov_l = 0
|
||||
for os in reversed(cur_sents):
|
||||
if ov_l + estimate_tokens(os) < overlap: ov_s.insert(0, os); ov_l += estimate_tokens(os)
|
||||
else: break
|
||||
cur_sents = list(ov_s); cur_sents.append(s); cur_len = ov_l + slen
|
||||
else: cur_sents.append(s); cur_len += slen
|
||||
if cur_sents: _add(" ".join(cur_sents), sec_title, sec_path)
|
||||
buf = []
|
||||
|
||||
for b in blocks:
|
||||
if b.kind == "heading": continue
|
||||
if estimate_tokens("\n\n".join([x.text for x in buf])) + estimate_tokens(b.text) >= target: flush()
|
||||
buf.append(b)
|
||||
if estimate_tokens(b.text) >= target: flush()
|
||||
flush()
|
||||
return chunks
|
||||
|
||||
def strategy_by_heading(blocks: List[RawBlock], config: Dict[str, Any], note_id: str, doc_title: str = "") -> List[Chunk]:
|
||||
"""Splittet Text basierend auf Markdown-Überschriften."""
|
||||
strict = config.get("strict_heading_split", False); target = config.get("target", 400)
|
||||
max_tokens = config.get("max", 600); split_level = config.get("split_level", 2)
|
||||
chunks = []; buf = []; cur_tokens = 0
|
||||
|
||||
def _flush(title, path):
|
||||
nonlocal buf, cur_tokens
|
||||
if not buf: return
|
||||
txt = "\n\n".join(buf); win = f"# {doc_title}\n## {title}\n{txt}".strip() if title else txt
|
||||
idx = len(chunks)
|
||||
chunks.append(Chunk(id=f"{note_id}#c{idx:02d}", note_id=note_id, index=idx, text=txt, window=win, token_count=estimate_tokens(txt), section_title=title, section_path=path, neighbors_prev=None, neighbors_next=None))
|
||||
buf = []; cur_tokens = 0
|
||||
|
||||
for b in blocks:
|
||||
if b.kind == "heading":
|
||||
if b.level < split_level: _flush(b.section_title, b.section_path)
|
||||
elif b.level == split_level:
|
||||
if strict or cur_tokens >= target: _flush(b.section_title, b.section_path)
|
||||
continue
|
||||
bt = estimate_tokens(b.text)
|
||||
if cur_tokens + bt > max_tokens and buf: _flush(b.section_title, b.section_path)
|
||||
buf.append(b.text); cur_tokens += bt
|
||||
if buf: _flush(blocks[-1].section_title if blocks else None, blocks[-1].section_path if blocks else "/")
|
||||
return chunks
|
||||
@@ -0,0 +1,55 @@
|
||||
"""
|
||||
FILE: app/core/chunking/chunking_utils.py
|
||||
DESCRIPTION: Hilfswerkzeuge für Token-Schätzung und YAML-Konfiguration.
|
||||
"""
|
||||
import math
|
||||
import yaml
|
||||
import logging
|
||||
from pathlib import Path
|
||||
from typing import Dict, Any, Tuple
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
BASE_DIR = Path(__file__).resolve().parent.parent.parent.parent
|
||||
CONFIG_PATH = BASE_DIR / "config" / "types.yaml"
|
||||
DEFAULT_PROFILE = {"strategy": "sliding_window", "target": 400, "max": 600, "overlap": (50, 80)}
|
||||
|
||||
_CONFIG_CACHE = None
|
||||
|
||||
def load_yaml_config() -> Dict[str, Any]:
|
||||
global _CONFIG_CACHE
|
||||
if _CONFIG_CACHE is not None: return _CONFIG_CACHE
|
||||
if not CONFIG_PATH.exists(): return {}
|
||||
try:
|
||||
with open(CONFIG_PATH, "r", encoding="utf-8") as f:
|
||||
data = yaml.safe_load(f)
|
||||
_CONFIG_CACHE = data
|
||||
return data
|
||||
except Exception: return {}
|
||||
|
||||
def get_chunk_config(note_type: str) -> Dict[str, Any]:
|
||||
"""Lädt die Chunking-Strategie basierend auf dem Note-Type."""
|
||||
full_config = load_yaml_config()
|
||||
profiles = full_config.get("chunking_profiles", {})
|
||||
type_def = full_config.get("types", {}).get(note_type.lower(), {})
|
||||
profile_name = type_def.get("chunking_profile") or full_config.get("defaults", {}).get("chunking_profile", "sliding_standard")
|
||||
config = profiles.get(profile_name, DEFAULT_PROFILE).copy()
|
||||
if "overlap" in config and isinstance(config["overlap"], list):
|
||||
config["overlap"] = tuple(config["overlap"])
|
||||
return config
|
||||
|
||||
def estimate_tokens(text: str) -> int:
|
||||
"""Grobe Schätzung der Token-Anzahl."""
|
||||
return max(1, math.ceil(len(text.strip()) / 4))
|
||||
|
||||
def extract_frontmatter_from_text(md_text: str) -> Tuple[Dict[str, Any], str]:
|
||||
"""Trennt YAML-Frontmatter vom Text."""
|
||||
import re
|
||||
fm_match = re.match(r'^\s*---\s*\n(.*?)\n---', md_text, re.DOTALL)
|
||||
if not fm_match: return {}, md_text
|
||||
try:
|
||||
frontmatter = yaml.safe_load(fm_match.group(1))
|
||||
if not isinstance(frontmatter, dict): frontmatter = {}
|
||||
except Exception: frontmatter = {}
|
||||
text_without_fm = re.sub(r'^\s*---\s*\n(.*?)\n---', '', md_text, flags=re.DOTALL)
|
||||
return frontmatter, text_without_fm.strip()
|
||||
Reference in new issue
Block a user