Oy 6 — MLOps va Production
🎯 Bu oydagi maqsad
**Bu oy — sizning asosiy maqsadingiz uchun eng muhim oy.**ML Engineer / MLOps Engineer bo'lish uchun shu oy bilim sizning portfoliongizning markazi bo'ladi.
Oy oxirida siz quyidagilarni qila olasiz:
- MLOps lifecycle'ni boshidan oxirigacha bilasiz
- MLflow bilan eksperimentlarni track qilasiz va modellarni versioning qilasiz
- DVC bilan data versioning va reproducibility ta'minlaysiz
- FastAPI + BentoML/TorchServe bilan ML modellarni serve qilasiz
- Docker + Kubernetes'da ML deployment
- Prometheus + Grafana + Evidently AI bilan monitoring va drift detection
- Apache Airflow bilan ML pipeline'larni orkestrlaysiz
- GitHub Actions bilan ML CI/CD
Haftalik taqsimot
| Hafta | Mavzu | Vaqt |
|---|---|---|
| Hafta 1 | MLOps intro + MLflow + DVC | 10-12 soat |
| Hafta 2 | FastAPI serving + Docker + K8s | 12-15 soat |
| Hafta 3 | Monitoring + CI/CD | 10-12 soat |
| Hafta 4 | Airflow + End-to-End capstone | 12-15 soat |
Boblar tartibi
- MLOps ga kirish
- MLflow — Experiment tracking
- DVC — Data Versioning
- FastAPI + ML Serving
- Docker va Kubernetes
- Model Monitoring
- CI/CD for ML
- Airflow va Prefect
- Mashqlar
Oy oxirida nima qila olasiz?
- To'liq production ML system qurish: training → versioning → serving → monitoring
- ML model deployment Kubernetes'da
- Drift detection bilan model degradation'ni avtomatik aniqlash
- CI/CD pipeline ML uchun (test, validate, deploy)
- Airflow DAG bilan haftalik retraining
- Job descriptionlarda yozilgan MLOps Engineertalablariga javob bera olish
Backend Dev uchun maslahat — bu oy sizning oltin oyingiz!
Sizning mavjud bilimlaringiz aynan shu oyda kuchli ustunlikberadi:
| Backend bilim | MLOps'da qo'llanish |
|---|---|
| Docker, docker-compose | ML containers |
| PostgreSQL | Feature store, prediction logs |
| Redis | Model cache, feature cache |
| Celery, Kafka | Async inference, streaming |
| GitHub Actions / GitLab CI | ML CI/CD |
| Nginx, load balancing | ML model serving |
| Prometheus, Grafana | ML monitoring |
| REST API design | ML inference endpoints |
| Async/await | Concurrent inference |
| Microservices | ML services architecture |
Aksariyat ML Engineerlar (data scientist'lardan kelganlar) bu narsalarni nol darajadano'rganishadi. Sizning boshlang'ich darajangizulardan ancha yuqori.
Cloud Cost (ixtiyoriy)
Bu oy uchun cloud xizmatlari kerak bo'ladi. Variantlar:
- AWS Free Tier($300 credit yangi accountlar)
- GCP Free Tier($300 credit)
- DigitalOcean($200 credit student/coupon)
- Hetzner — eng arzon (€5/oy server)
- Lokal Kubernetes(minikube, kind, k3s) — bepul, kichik loyihalar uchun yetadi
**Maslahat:**Asosiy mashqlar lokal Docker + minikube bilan, faqat capstone uchun real cloud.
Boshlash
MLOps ga kirish bilan boshlang.