Chapter 34: Introduction to MLOps
MODULE 6 | MLOps
Chapter 34: Introduction to MLOps
Ye chapter poore module ki bunyad hai. Hum dekhenge ke MLOps kya hai , notebook wala model production me kyun nahi chalta, ek ML project ki poori zindagi (lifecycle) kaisi hoti hai, aur is module ke tools kis masle ka hal hain. Aakhir me model ko save aur load karna seekhenge, kyunke deployment ka pehla qadam yehi hai.
34.1 MLOps kya hai aur kyun zaroori hai?
MLOps (Machine Learning Operations) un tareeqon aur tools ka naam hai jin se ML model ko bharosemand tareeqe se banaya, chalaya aur sambhala jata hai. Ye teen duniyaon ka milaap hai: Machine Learning (model banana), DevOps (software ko bina masle ke chalana) aur Data Engineering (data ko saaf aur tayyar rakhna).
34.2 Notebook aur Production me farq
| Cheez | Notebook (tumhara computer) | Production (asli duniya) |
|---|---|---|
| Kaun use karta hai | Sirf tum | Hazaron log, apps, doosre systems |
| Model kaise chalta hai | Cell chalao, jawab dekho | API ke zariye,24 ghante, khud ba khud |
| Environment | Tumhari library versions | Har jagahek jaisahona chahiye (Docker) |
| Galti hone par | Cell dobara chalao | Log dekho, alert aaye, rollback karo |
| Model purana hone par | Pata hi nahi chalta | Monitoringse pata chalta hai |
| Experiments ka record | Yaad ya notebook ke comments | MLflowme parameters aur metrics |
Sab se mashhoor jumla: "Mere computer par to chal raha tha." MLOps ka maqsad ye jumla khatam karna hai.
34.3 ML Lifecycle
Ek ML project ek baar banta aur khatam nahi hota . Wo ek chakkar (cycle) me chalta rehta hai. Ye rahe is ke marhale:
DATA -> TRAINING -> EVALUATION -> DEPLOYMENT -> MONITORING
^ |
|________________ naya data, naya model _________________|
| Marhala | Kya hota hai | Is module me tool |
|---|---|---|
| Data | Data jama karna, saaf karna, version rakhna | Module 1 se 4 (pandas, sklearn) |
| Training | Model train karna, experiments ka record | MLflow(Chapter 37) |
| Evaluation | Test data par accuracy dekhna, models compare karna | MLflow UI, pytest (Chapter 37, 38) |
| Deployment | Model ko API bana kar chalana, dabbe me band karna | FastAPI,Docker(Chapter 35, 36) |
| Automation | Har badlav par khud test aur build | GitHub Actions(Chapter 38) |
| Monitoring | Live model ki sehat, log, data drift | logging, drift check (Chapter 39) |
34.4 Pehla qadam: Model save aur load karna
Model ko API ya Docker me chalane se pehle usay file me save karna hota hai. Warna har baar training dobara karni parti. scikit-learn ke models ke liye sab se aam tareeqa joblib hai:
import joblib
from sklearn.datasets import load_iris
from sklearn.ensemble import RandomForestClassifier
X, y = load_iris(return_X_y=True)
model = RandomForestClassifier(n_estimators=50, random_state=42)
model.fit(X, y)
joblib.dump(model, "iris_model.joblib")
print("Model save ho gaya")
loaded = joblib.load("iris_model.joblib")
sample = [[5.1, 3.5, 1.4, 0.2]]
print("Prediction:", loaded.predict(sample)[0])
OUTPUT
Model save ho gaya
Prediction: 0
Prediction 0 ka matlab setosa hai (Iris ki pehli qism). Dhyan do: model file ek hi baar train hoti hai , aur uske baad API ya Docker usay sirf load karte hain.
AAM GALTIYAN
-
Un jaanchi hui file load karna:
joblibaurpicklefiles me code chhup sakta hai. Sirf wo model file load karo jo tum ne khud banayi ya jis par tumhe poora bharosa hai. -
Version mismatch: jis scikit-learn version se model train kiya, deployment me wohi version install karo, warna error ya galat nateeja aa sakta hai. Isi liye
requirements.txtme versions likhte hain.
34.5 Project ka setup: folder aur virtual environment
Har project ke liye alag virtual environment banao taake libraries ek doosre se na takrayen. Terminal me ye karo:
TERMINAL
mkdir mlops-practice
cd mlops-practice
python -m venv venv
# Windows:
venv\Scripts\activate
# Mac / Linux:
source venv/bin/activate
pip install fastapi uvicorn scikit-learn joblib pandas httpx pytest mlflow
pip freeze > requirements.txt
VERSION NOTE
Is module me Python 3.12 use hua hai. Bohat naye Python versions (maslan 3.14) par kuch libraries (jese TensorFlow ) abhi support nahi karti, aur nayi library versions me chhoti chhoti tabdeeliyan aa sakti hain. Agar koi code error de, to pehle python --version aur pip list dekho. Is module me FastAPI 0.159, scikit-learn 1.8, MLflow 3.17, pandas 3.0 ke sath test kiya gaya hai.
34.6 Aam Galtiyan aur Hands-on
AAM GALTIYAN
-
MLOps ko sirf "Docker seekhna" samajhna. Docker iska ek hissa hai; asal cheez poora lifecycle hai.
-
Model deploy kar ke bhool jana . Live model ko nigrani aur dobara training chahiye.
-
Notebook ko hi production samajhna. Production code
.pyfiles, tests aur config me hota hai. -
Environment ka record na rakhna (
requirements.txtna banana).
Hands-on
-
Apne kisi purane project ka model
joblibse save karo aur naye notebook me load kar ke ek prediction karo. -
mlops-practicefolder aur virtual environment banao, aurpip freeze > requirements.txtchalao. -
Apne kisi purane project ke liye likho: wo lifecycle ke kaun se marhale tak pohancha hai (data, training, deployment, monitoring)?
-
Ek jumle me likho: "Mere computer par chalta hai" production me kyun kaafi nahi?
KEY TAKEAWAYS
-
MLOps = ML + DevOps + Data Engineering: model ko bharosemand tareeqe se banana, chalana aur sambhalna.
-
Notebook me model sirf tum chalate ho; production me API, 24 ghante, ek jaisa environment chahiye.
-
ML lifecycle ek chakkar hai: Data, Training, Evaluation, Deployment, Monitoring, phir wapas Data.
-
joblib.dump/joblib.loadse model save aur load hota hai; sirf bharose ki files load karo. -
requirements.txtme versions likho taake har jagah wohi environment bane.