Table of contents
Open Table of contents
1. Dinh nghia
MlOps la viec kiem soat va van hanh cac hoat dong de duocm otsan pham ML hooan chinh tu train, validate, deploy va monitor
Kiem soat o day la chung ta kiem soat du an mot cac de du doan nhat co the va theo mot he thong voi muc dich la co the tai san xuat mot cach de dang.
1.1. Use cases
1.1.1. Model drift
Gia su model chung ta dang duoc dung voi muc dich la spam detection, tuy nhien la chung ta muon chuyen mo hinh qua fraud detection. Nhu the thi neu chung ta thuc hien MLOps mot cach dung dan thi qua trinh nay se tro nen de dang hon.
1.1.2. Data drift
Tuong tu voi model drift, khi chung ta nhan thay du lieu dau vao de train model khong con thich hop nua va chung ta muon doi du lieu de train thi voi MLOps qua trinh ay co the dien ra mot cach nhanh chong
1.2. Need for MLOps:
- Experimentation
- Track metrics
- Source control the code
- Checkpoint steps in the ML lifecycle
- Automating proper validation/staged deployment
- Monitoring model performacnce efficiency + automated retraining
2. Frameworks
2.1. MlFlow vs CleanML
MLflow cung cap end-to-end cycle cho ML. Trong do, nen tang se log lai nhung thong tin cac metric sau moi epoch cung nhu la cac parameter. Ngoai ra MLflow con ho tro dong goi model va deploying model.