Oy 2 — Klassik ML (Scikit-learn)

🎯 Bu oydagi maqsad

Oy oxirida siz quyidagilarni qilolasiz:

• Real biznes muammosini ML masalasiga aylantirish (regression / classification / clustering)
• Scikit-learn bilan to'liq ML pipeline qurish
• Modelni baholash va xato turlarini tushunish
• XGBoost/LightGBM bilan production darajadagi modellar yaratish
• Kaggle competition'da qatnashish va top 30%'ga kirish

Haftalik taqsimot

Boblar tartibi

1. ML ga kirish — terminlar, jarayon, training/test split
2. Regression — uzluksiz qiymatni bashorat qilish
3. Classification — sinflarga ajratish
4. Clustering — unsupervised guruhlash
5. Feature Engineering — feature'larni tayyorlash va yaratish
6. Model Evaluation — metrik va validation
7. Ensemble Methods — Random Forest, XGBoost, LightGBM
8. Mashqlar — qo'shimcha mashqlar va Kaggle topshiriqlari

Oy oxirida nima qila olasiz?

• Tabular ma'lumotlarda 80% problemalarni hal qilish (regression, classification, clustering)
• Scikit-learn Pipeline yordamida reproducible kod yozish
• Modelni joblib bilan saqlash va FastAPI'da serve qilish
• XGBoost/LightGBM bilan Kaggle'da top 30%
• ML modelining biznes uchun ROIni tushuntirish

Backend Dev uchun maslahat

Backend'da REST API yozish kabi, ML'da fit() → predict() pattern bor:

# Backend pattern
@app.post("/users")
def create_user(data: UserIn):
    user = User(**data.dict())
    db.add(user)
    return user

# ML pattern
model = LogisticRegression()
model.fit(X_train, y_train)        # "train" qilish
predictions = model.predict(X_test) # "predict" qilish

Bu pattern barcha sklearn modellarda bir xil — agar LinearRegressionni bilsangiz, RandomForestni ham bilasiz.

MLOps integration (boshidan)

Backend dev'ning ustunligi — production thinking. Birinchi modelni yozayotganingizda allaqachon o'ylang:

1. Reproducibility — random_state=42 har joyda
2. Saqlash — joblib.dump(model, 'model.pkl')
3. Versioning — model_v1.pkl, model_v2.pkl
4. Schema — Pydantic bilan input/output validation
5. Logging — har bashorat uchun timestamp va input

Bu mavzular Oy 6 (MLOps)'da chuqurroq, lekin birinchi kunidan boshlang.

Boshlash

ML ga kirish bilan boshlang.