How koverthanan-git/credit_card_ct_project works
This repository implements a credit-card fraud ML workflow: an operator can train and register a model, while a Prefect flow checks for data drift and conditionally retrains it. A FastAPI inference gateway loads the latest registered model and returns fraud predictions. The README also describes CI/CD and evaluation, but the excerpts do not show a CI/CD implementation; the sampled evaluation is model accuracy during initial training. The graph reflects the visible calls and data transfers; configuration files are not connected because their runtime wiring is not shown.
Main entry points: train.py Initial training · pipline.py Training flow · monitor.py Drift evaluation · app.py Inference gateway
Components 11
Model lifecycle
- Initial training
train.py - Reference data
train.py - MLflow registry
Drift and retraining
- Training flow
pipline.py - Drift check
pipline.py - Drift evaluation
monitor.py - Conditional retraining
pipline.py - Drift dashboard
monitor.py
Inference serving
- Inference gateway
app.py
Other components
- Operator
- Inference client
Connections 13
- Operator → Initial training: starts training
- Initial training → Reference data: writes baseline
train.py - Initial training → MLflow registry: registers model
train.py - Operator → Training flow: starts flow
- Training flow → Drift check: runs task
pipline.py - Drift check → Drift evaluation: calls evaluator
pipline.py - Drift evaluation → Reference data: reads baseline
monitor.py - Drift evaluation → Drift dashboard: saves report
monitor.py - Training flow → Conditional retraining: dispatches on drift
pipline.py - Conditional retraining → MLflow registry: registers model
pipline.py - Inference client → Inference gateway: submits transaction
- Inference gateway → MLflow registry: loads latest model
app.py - Inference gateway → Inference client: returns prediction
app.py