Lesson 01 � Foundation

MACHINE LEARNING
KYA HAI?

Spam mail automatically filter ho jaata hai � kaise✓ Netflix ko kaise pata hai ki aapko kaunsa show pasand aayega✓ Self-driving cars road kaise samajhti hain✓ Yeh sab Machine Learning ki wajah se hota hai. ML mein computer khud se data se patterns seekhta hai aur predictions deta hai.

? 18 min✓ Beginner✓ No Prerequisite

WHY: Machine Learning kyun important hai?

Aaj ke time mein har jagah ML hai � jab WhatsApp pe spam message aata hai aur woh automatically filtered ho jaata hai, jab Amazon pe kuch search karte ho aur "Customers also bought" dikhta hai, jab PhonePe fraud detect karta hai � yeh sab ML ke examples hain. ML se computers itne smart ho gaye hain ki woh khud se decisions le sakte hain bina har baar human ko bolne ke.

SUPERVISED

Labeled data se seekhna. Jaise 1000 emails diye jinke labels hain "spam" ya "not spam" � phir model naye email ko classify karta hai. Labeled data ka matlab: input ke saath sahi jawab bhi diya hota hai.

UNSUPERVISED

Bina labels ke patterns dhundhna. Jaise 500 customers ka data ho aur model khud se groups bana de � "yeh log saste products khareedte hain" ya "yeh log premium users hain." Koi sahi jawab nahi diya hota, model khud dhoondhta hai.

REINFORCEMENT

Rewards aur penalties se seekhna. Jaise game AI khelna seekhta hai � sahi move pe reward milta hai, galat move pe penalty.✓ better hota jaata hai. AlphaGo, chess AI � sab reinforcement learning ke examples hain.

ALGORITHM

Rules for learning. Har ML algorithm ek specific tarika hai data se patterns seekhne ka. Linear Regression, Decision Trees, Neural Networks � sab algorithms hain. Problem ke hisaab se algorithm choose karte hain.

WHAT: ML exactly kya karta hai?

Simple terms mein: ML ek aisa process hai jismein computer ko data dete hain aur woh usse patterns seekhta hai. Phir woh patterns use karke naye data pe predictions karta hai. Yeh traditional programming se alag hai � traditional mein hum rules likhte hain, ML mein computer khud rules dhoondhta hai.

Traditional Programming vs Machine Learning:

TRADITIONAL PROGRAMMING:
 Input: Data + Rules
 Output: Answers

 Example: Email spam filter banana
 ✓ Rules likho: agar subject mein "FREE" hai toh spam
 ✓ Problem: Har rule kaun likhega✓ Bahut complex ho jaata hai

MACHINE LEARNING:
 Input: Data + Answers (labels)
 Output: Rules (model)

 Example: Email spam filter banana
 ? 10,000 emails do jinke labels hain (spam/not spam)
 ✓ Model khud se rules dhoondh leta hai
 ✓ Ab naye email ko automatically classify karta hai

ℹ

Mental model: ML jaise bachche ka seekhna hai � aap bachche ko 10 baar apple dikhao aur bolo "yeh apple hai", phir 11vi baar naya fruit dikhao toh woh khud se identify kar lega. Computer bhi isi tarah data se seekhta hai.

HOW: ML ke 3 main types

Machine Learning ko 3 broad categories mein baanta hai. Har category ka apna tareeka hai data se seekhne ka. Yeh samajhna zaroori hai kyunki problem aayegi tab aapko pata hona chahiye kaunsa type use karna hai.

1. SUPERVISED LEARNING (Labeled data se seekhna):
 
 Data mein dono hote hain: input + sahi jawab
 
 Examples:
 ✓ Spam Detection: Emails + labels (spam/not spam)
 ✓ House Price Prediction: Size + location + price
 ✓ Medical Diagnosis: Symptoms + disease name
 
 Sub-types:
 ✓ Classification: Discrete categories (spam/not spam)
 ✓ Regression: Continuous values (price, temperature)


2. UNSUPERVISED LEARNING (Bina labels ke patterns):
 
 Data mein sirf input hai, koi sahi jawab nahi
 
 Examples:
 ✓ Customer Segmentation: Customers ko groups mein baanto
 ✓ Anomaly Detection: Unusual transactions dhundho
 ✓ Market Basket Analysis: "Yeh 2 products saath khareede jaate hain"
 
 Sub-types:
 ✓ Clustering: Groups banana (K-Means)
 ✓ Dimensionality Reduction: Features kam karna (PCA)


3. REINFORCEMENT LEARNING (Rewards se seekhna):
 
 Agent environment mein action leta hai, reward milta hai
 
 Examples:
 ✓ Game AI: Chess, Go, Video games
 ✓ Robotics: Robot ko chalna seekhana
 ✓ Self-Driving Cars: Road pe decisions lena
 
 Key terms:
 ✓ Agent: Jo seekh raha hai
 ✓ Environment: Jismein woh kaam kar raha hai
 ✓ Reward: Sahi kaam pe bonus

HOW: ML Pipeline � 6 steps

ML project mein step by step chalte hain. Yeh 6 steps yaad rakho � yeh har ML project mein apply honge. Galat step pe galat result aata hai.

ML ka 6-step pipeline:

1. DATA COLLECT
 ✓ Data uthao � CSV, database, API, ya web scraping se
 ✓ Jitna zyada data, utna better model
 ✓ Example: 10,000 house prices ka dataset

2. DATA CLEAN
 ✓ Missing values handle karo
 ✓ Duplicates hatao
 ✓ Outliers identify karo
 ✓ Data types fix karo

3. FEATURE SELECTION
 ✓ Kaunse columns important hain?
 ✓ Irrelevant features hatao
 ✓ New features banao (feature engineering)
 ✓ Example: "Year Built" se "Age" nikalna

4. MODEL TRAIN
 ✓ Algorithm choose karo
 ✓ Training data se model ko teach karo
 ✓ Model patterns seekhta hai
 ✓ Parameters adjust hote hain

5. EVALUATE
 ✓ Test data pe check karo model kaisa perform karta hai
 ✓ Metrics dekho: Accuracy, Precision, Recall
 ✓ Overfitting ya underfitting toh nahi?

6. DEPLOY
 ✓ Real-world mein use karo
 ✓ API banao ya dashboard lagao
 ✓ Monitor karo model abhi bhi sahi kaam kar raha hai ya nahi

HOW: real-world ML examples

ML sirf tech companies ka nahi � aapki daily life mein bhi hai. Har baar jab smart features use karte ho, ML kaam kar raha hai.

ML in Daily Life:

Entertainment:
 ✓ Netflix/Amazon: Recommendation engine
 ✓ YouTube: Video suggestions
 ✓ Spotify: Discover Weekly playlist

Technology:
 ✓ Gmail: Spam detection
 ✓ Google Search: PageRank algorithm
 ✓ Siri/Alexa: Voice recognition
 ✓ Google Translate: Language translation

Finance:
 ✓ Credit card fraud detection
 ✓ Stock price prediction
 ✓ Loan approval decisions

Healthcare:
 ✓ Disease prediction from symptoms
 ✓ Medical image analysis (X-ray, MRI)
 ✓ Drug discovery

Transport:
 ✓ Self-driving cars (Tesla, Waymo)
 ✓ Route optimization (Google Maps)
 ✓ Uber/Lyft ride matching

Shopping:
 ✓ Product recommendations
 ✓ Dynamic pricing
 ✓ Customer behavior prediction

Try it: Python se ML start

Neeche ka editor Python jaisa hai. Yahan chhota sa ML code likho aur "Run Python" dabao. Screen par output dikhenge � yeh browser-based execution hai, real Python chalega.

Python playgroundML ka pehla kadam � Linear Regression

main.py

# Machine Learning example: House Price Prediction
# Simple Linear Regression with scikit-learn

import numpy as np

# Step 1: Data collect karo
# House size (sq ft) vs Price (lakhs)
X = np.array([600, 800, 1000, 1200, 1400, 1600, 1800, 2000]).reshape(-1, 1)
y = np.array([25, 30, 38, 42, 50, 58, 65, 72])

print("=== House Price Data ===")
print(f"Houses: {len(X)}")
for size, price in zip(X.flatten(), y):
 print(f" {size} sq ft -> {price} lakhs")

# Step 2: Simple Linear Regression manually
# y = mx + b (slope-intercept form)
x_mean = np.mean(X)
y_mean = np.mean(y)

# Calculate slope (m) and intercept (b)
numerator = np.sum((X.flatten() - x_mean) * (y - y_mean))
denominator = np.sum((X.flatten() - x_mean) ** 2)
m = numerator / denominator
b = y_mean - m * x_mean

print(f"\n=== Model Trained ===")
print(f"Formula: Price = {m:.4f} x Size + {b:.2f}")
print(f"Slope (m): {m:.4f} � Har 1 sq ft badhne pe price {m:.4f} lakhs badhta hai")

# Step 3: Predict karo
new_sizes = [900, 1500, 2200]
print(f"\n=== Predictions ===")
for size in new_sizes:
 predicted = m * size + b
 print(f" {size} sq ft -> {predicted:.2f} lakhs")

# Step 4: Model evaluate karo (simple R-squared)
y_pred = m * X.flatten() + b
ss_res = np.sum((y - y_pred) ** 2)
ss_tot = np.sum((y - y_mean) ** 2)
r_squared = 1 - (ss_res / ss_tot)
print(f"\n=== Model Accuracy ===")
print(f"R-squared: {r_squared:.4f}")
print(f"{'Excellent' if r_squared > 0.9 else 'Good' if r_squared > 0.7 else 'Needs work'} fit!")

Optional input

Code ko apni info se update karke run karein

Quick check

Supervised aur Unsupervised mein 3 differences batao.

Sochho � supervised mein teacher hota hai jo sahi jawab batata hai (labeled data), unsupervised mein student ko khud se patterns dhundhne hote hain. Supervised mein output known hai, unsupervised mein unknown.

Common beginner mistakes

Data collection skip karna: ML 80% data cleaning hai. Agar data saaf nahi hai toh model bhi galat seekhega. Pehle data pe time lagao.
Sab algorithms ek saath try karna: Pehle ek algorithm samjho (Linear Regression), phir doosre pe jao. Ek hi problem pe 10 algorithms mat lagao.
Training aur test data mix karna: Model ko woh data mat do jo usne pehle dekha hai. Hamesha test data alag rakho � warna accuracy fake aayegi.
Results pe mat jaano, process pe jaano: Kaggle leaderboard nahi, problem-solving approach matter karta hai. Step by step samjho kya ho raha hai.

ML Introduction clear?

Ab Supervised Learning par chalo � labeled data se kaise seekhte hain. Yeh ML ka sabse common type hai.

MACHINE LEARNINGKYA HAI?