marcel/familienarchiv
1
0
Fork 0
Code Issues 119 Pull Requests 2 Actions Packages Projects Releases Wiki Activity

feat(training): track and display CER per training run

Browse Source 
After each training run, the Character Error Rate (CER = 1 - accuracy),
loss, accuracy, and epoch count are now stored on the OcrTrainingRun
record and shown in the training history table.

Also adds the missing POST /api/ocr/segtrain endpoint and the
triggerSegTraining service method so the segmentation training card
can actually trigger training.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Marcel
2026-04-13 19:01:10 +02:00
parent a99afef319
commit 22954f348a
12 changed files with 118 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
ocr-service/main.py
View File

@@ -333,8 +333,10 @@ async def train_model(
epochs = getattr(result, "epochs", None) or 0
loss = getattr(result, "best_loss", None)
accuracy = getattr(result, "best_accuracy", None)
# CER = 1 - character accuracy (Kraken reports character-level accuracy)
cer = round(1.0 - accuracy, 4) if accuracy is not None else None
log.info("Training complete — epochs=%s loss=%s accuracy=%s", epochs, loss, accuracy)
log.info("Training complete — epochs=%s loss=%s accuracy=%s cer=%s", epochs, loss, accuracy, cer)
# Backup existing model and replace
if os.path.exists(KRAKEN_MODEL_PATH):
@@ -351,7 +353,7 @@ async def train_model(
kraken_engine.load_models()
log.info("Reloaded Kraken model in-process")
return {"loss": loss, "accuracy": accuracy, "epochs": epochs}
return {"loss": loss, "accuracy": accuracy, "cer": cer, "epochs": epochs}
result = await asyncio.to_thread(_run_training)
return result
@@ -409,7 +411,7 @@ async def segtrain_model(
loss = getattr(result, "best_loss", None)
accuracy = getattr(result, "best_accuracy", None)
log.info("Segmentation training complete — epochs=%s loss=%s", epochs, loss)
log.info("Segmentation training complete — epochs=%s loss=%s accuracy=%s", epochs, loss, accuracy)
if os.path.exists(blla_model_path):
timestamp = datetime.now(timezone.utc).strftime("%Y%m%dT%H%M%SZ")
@@ -420,7 +422,7 @@ async def segtrain_model(
shutil.move(output_model_path, blla_model_path)
log.info("Replaced blla model at %s", blla_model_path)
return {"loss": loss, "accuracy": accuracy, "epochs": epochs}
return {"loss": loss, "accuracy": accuracy, "cer": None, "epochs": epochs}
result = await asyncio.to_thread(_run_segtrain)
return result