bank-statement-api / routers /classification.py
pratik-befree's picture
Update classification.py
e16fc34
Raw History Blame Contribute Delete
14.1 kB
import os
import uuid
import logging
from typing import Optional, List
from fastapi import APIRouter, Depends, HTTPException, BackgroundTasks, UploadFile, File
import pandas as pd
from starlette.concurrency import run_in_threadpool
from auth.models import User
from auth.service import get_current_user
from models.admin_models import FinalisePayload
from models.classification import CategoryOverride
from models.transactions import BulkConfirmPayload
from models.statements import UploadInitResponse
from core.state import statements, settings, StatementData
from core.helpers import _filter_transactions_by_last_n_months, _normalize_description
from core.statement_io import _persist_swap_to_xlsx
from core.pipeline import get_classification_pipeline
logger = logging.getLogger("befree.classification")
router = APIRouter(prefix="/api/v1", tags=["classification"])
async def run_statement_classification_task(statement_id: str, extract_last_3_months: Optional[bool] = None):
"""Internal helper to run the classification pipeline on a statement in the background."""
try:
import logging as std_logging
logger = std_logging.getLogger("classification")
logger.setLevel(std_logging.INFO)
if not logger.handlers:
ch = std_logging.StreamHandler()
ch.setFormatter(std_logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s'))
logger.addHandler(ch)
if not os.environ.get("GOOGLE_API_KEY"):
logger.warning("GOOGLE_API_KEY not found in environment. LLM classification will be skipped.")
stmt = statements[statement_id]
if not stmt.transactions:
logger.error(f"No transactions to classify for {statement_id}")
return {"error": "No transactions to classify"}
use_3_months = extract_last_3_months if extract_last_3_months is not None else settings.extractLast3Months
stmt.extract_last_3_months = use_3_months
classify_candidates = stmt.transactions
if use_3_months:
classify_candidates = _filter_transactions_by_last_n_months(stmt.transactions)
logger.info(
f"Starting classification for statement {statement_id} "
f"(extract_last_3_months={use_3_months}, "
f"{len(classify_candidates)}/{len(stmt.transactions)} transactions selected for classification)"
)
try:
from overwatch.event_log import EventLog as _OWEventLog
_ow_cls_el = _OWEventLog(document_id=statement_id, document_name=stmt.file_name)
except Exception:
_ow_cls_el = None
pipeline = get_classification_pipeline()
if _ow_cls_el is not None:
try:
pipeline._event_log = _ow_cls_el
except Exception:
pass
classified_subset = await run_in_threadpool(
pipeline.classify_batch, classify_candidates
)
try:
pipeline._event_log = None
except Exception:
pass
for tx in classified_subset:
if "category_confidence" in tx:
tx["categoryConfidence"] = tx.pop("category_confidence")
if "category_source" in tx:
tx["categorySource"] = tx.pop("category_source")
if "third_party" in tx:
tx["thirdParty"] = tx.pop("third_party")
if "classification_reason" in tx:
tx["classificationReason"] = tx.pop("classification_reason")
classified_by_id = {tx.get("id"): tx for tx in classified_subset}
merged_transactions = []
for tx in stmt.transactions:
tx_id = tx.get("id")
if tx_id and tx_id in classified_by_id:
merged_transactions.append({**tx, **classified_by_id[tx_id]})
else:
merged_transactions.append(tx)
stmt.transactions = merged_transactions
stmt.status = "completed"
logger.info(f"Finished classification for statement {statement_id}")
try:
_persist_swap_to_xlsx(statement_id)
except Exception as e:
logger.error(f"Failed to persist classified statement {statement_id}: {e}")
counts = {"classified": 0, "rule": 0, "ml": 0, "llm": 0, "fallback": 0, "uncategorized": 0}
for tx in stmt.transactions:
raw_src = tx.get("categorySource") or tx.get("category_source") or ""
src = raw_src.split(":")[0] if raw_src else "uncategorized"
if src in ("rule", "ml", "llm", "manual"):
counts["classified"] += 1
if src in counts:
counts[src] += 1
elif src == "fallback":
counts["fallback"] += 1
else:
counts["uncategorized"] += 1
return counts
except Exception as exc:
logger.error(f"Classification failed for {statement_id}: {exc}", exc_info=True)
if statement_id in statements:
statements[statement_id].status = "failed"
return {"error": "Classification failed"}
@router.post("/statements/{statement_id}/finalise")
async def finalise_statement(
statement_id: str,
background_tasks: BackgroundTasks,
payload: FinalisePayload = FinalisePayload(),
current_user: User = Depends(get_current_user),
):
"""Lock the statement and trigger classification pipeline."""
if statement_id not in statements:
raise HTTPException(status_code=404, detail="Statement not found")
stmt = statements[statement_id]
stmt.status = "classifying"
use_3_months = payload.extract_last_3_months if payload.extract_last_3_months is not None else settings.extractLast3Months
stmt.extract_last_3_months = use_3_months
background_tasks.add_task(run_statement_classification_task, statement_id, use_3_months)
return {"status": "classifying", "message": "Classification started in background"}
@router.post("/statements/{statement_id}/classify")
async def classify_statement(statement_id: str, current_user: User = Depends(get_current_user)):
"""Run the classification pipeline on all transactions."""
res = await run_statement_classification_task(statement_id)
if isinstance(res, dict) and "error" in res:
raise HTTPException(status_code=500, detail=res["error"])
return res
@router.get("/statements/{statement_id}/classify/status")
async def classify_status(statement_id: str, current_user: User = Depends(get_current_user)):
"""Get classification status for a statement."""
if statement_id not in statements:
raise HTTPException(status_code=404, detail="Statement not found")
stmt = statements[statement_id]
if settings.extractLast3Months:
classify_candidates = _filter_transactions_by_last_n_months(stmt.transactions)
else:
classify_candidates = list(stmt.transactions)
total = len(classify_candidates)
breakdown = {"rule": 0, "ml": 0, "llm": 0, "manual": 0, "uncategorized": 0}
for tx in classify_candidates:
raw_src = tx.get("categorySource") or tx.get("category_source") or ""
src = raw_src.split(":")[0] if raw_src else "uncategorized"
if src in breakdown:
breakdown[src] += 1
elif src:
breakdown.setdefault(src, 0)
breakdown[src] += 1
else:
breakdown["uncategorized"] += 1
classified = total - breakdown.get("uncategorized", 0)
return {
"total": total,
"classified": classified,
"breakdown": breakdown,
"status": stmt.status
}
@router.patch("/statements/{statement_id}/classify/{tx_id}")
async def override_classification(
statement_id: str,
tx_id: str,
payload: CategoryOverride,
background_tasks: BackgroundTasks,
current_user: User = Depends(get_current_user),
):
"""Manually override a transaction's category."""
if statement_id not in statements:
raise HTTPException(status_code=404, detail="Statement not found")
stmt = statements[statement_id]
for tx in stmt.transactions:
if tx.get("id") == tx_id:
tx["category"] = payload.category
tx["parentCategory"] = payload.category
tx["group_name"] = payload.category
tx["groupName"] = payload.category
tx["sub_category"] = None
tx["subCategory"] = None
tx["categoryConfidence"] = "manual"
tx["category_confidence"] = "manual"
tx["categorySource"] = "manual"
tx["category_source"] = "manual"
background_tasks.add_task(_persist_swap_to_xlsx, statement_id)
return {"status": "success", "transaction": tx}
raise HTTPException(status_code=404, detail="Transaction not found")
@router.post("/statements/{statement_id}/classify/{tx_id}/confirm")
async def confirm_classify_suggestion(
statement_id: str,
tx_id: str,
background_tasks: BackgroundTasks,
current_user: User = Depends(get_current_user),
):
"""Verify an AI-suggested classification."""
if statement_id not in statements:
raise HTTPException(status_code=404, detail="Statement not found")
stmt = statements[statement_id]
target_tx = None
for tx in stmt.transactions:
if tx.get("id") == tx_id:
target_tx = tx
break
if target_tx is None:
raise HTTPException(status_code=404, detail="Transaction not found")
target_desc = target_tx.get("description", "")
target_cat = target_tx.get("category")
verified_ids: List[str] = []
for tx in stmt.transactions:
raw_src = tx.get("categorySource") or ""
is_ai = raw_src.startswith("ml") or raw_src.startswith("llm")
if not is_ai:
continue
if tx.get("description", "") == target_desc and tx.get("category") == target_cat:
tx["verified"] = True
verified_ids.append(tx["id"])
if tx_id not in verified_ids:
target_tx["verified"] = True
verified_ids.append(tx_id)
if verified_ids:
background_tasks.add_task(_persist_swap_to_xlsx, statement_id)
bulk_applied = len(verified_ids) > 1
logger.info(
f"[ConfirmClassify] {statement_id}: verified {len(verified_ids)} tx(s) "
f"matching '{target_desc}' (bulk={bulk_applied})"
)
return {
"success": True,
"verifiedIds": verified_ids,
"count": len(verified_ids),
"bulkApplied": bulk_applied,
}
@router.post("/statements/{statement_id}/classify/confirm-bulk")
async def confirm_classify_suggestions_bulk(
statement_id: str,
payload: BulkConfirmPayload,
background_tasks: BackgroundTasks,
current_user: User = Depends(get_current_user),
):
"""Bulk-verify a specific list of transaction IDs."""
if statement_id not in statements:
raise HTTPException(status_code=404, detail="Statement not found")
stmt = statements[statement_id]
tx_id_set = set(payload.txIds)
verified_ids: List[str] = []
for tx in stmt.transactions:
if tx.get("id") in tx_id_set:
tx["verified"] = True
verified_ids.append(tx["id"])
if verified_ids:
background_tasks.add_task(_persist_swap_to_xlsx, statement_id)
logger.info(f"[ConfirmBulk] {statement_id}: verified {len(verified_ids)} tx(s)")
return {
"success": True,
"verifiedIds": verified_ids,
"count": len(verified_ids),
}
@router.delete("/statements/{statement_id}/classify/{tx_id}/override")
async def reset_classification_override(
statement_id: str,
tx_id: str,
background_tasks: BackgroundTasks,
current_user: User = Depends(get_current_user),
):
"""Reset a transaction's manual category override."""
if statement_id not in statements:
raise HTTPException(status_code=404, detail="Statement not found")
stmt = statements[statement_id]
for tx in stmt.transactions:
if tx.get("id") == tx_id:
tx["category"] = None
tx["categoryConfidence"] = "unclassified"
tx["category_confidence"] = "unclassified"
tx["categorySource"] = "none"
tx["category_source"] = "none"
background_tasks.add_task(_persist_swap_to_xlsx, statement_id)
return {"status": "success", "transaction": tx}
raise HTTPException(status_code=404, detail="Transaction not found")
@router.post("/classify/upload-excel")
async def upload_excel_for_classification(
background_tasks: BackgroundTasks,
file: UploadFile = File(...),
current_user: User = Depends(get_current_user)
):
"""Upload an Excel file directly for classification."""
try:
content = await file.read()
df = pd.read_excel(content)
statement_id = str(uuid.uuid4())
transactions = []
for idx, row in df.iterrows():
tx = {
"id": str(uuid.uuid4()),
"date": str(row.get("Date", "")),
"description": str(row.get("Description", "")),
"amount": float(row.get("Amount", 0)),
"category": None,
"categoryConfidence": "unclassified",
}
if tx["amount"] < 0:
tx["debit"] = abs(tx["amount"])
tx["credit"] = None
else:
tx["credit"] = tx["amount"]
tx["debit"] = None
transactions.append(tx)
stmt = StatementData(statement_id, file.filename)
stmt.status = "classifying"
stmt.transactions = transactions
statements[statement_id] = stmt
background_tasks.add_task(run_statement_classification_task, statement_id)
return {"statementId": statement_id}
except Exception as e:
raise HTTPException(status_code=500, detail=str(e))