1. Question 1

What is wrong with the code?

• ❌ “LR” not required

• ❌ nn.Module not required

• ❌ super not required

• ✅ “linear” should be self.linear

Explanation:
Inside forward, PyTorch modules must be referenced using self.linear, not linear.

---

2. Question 2

Noise in linear regression refers to:

• ❌ Variation in parameters

• ✅ Random errors added to data points

• ❌ Lack of linearity

• ❌ Data collection errors

Explanation:
Noise = unavoidable random variability in real-world data.

---

3. Question 3

Purpose of Mean Squared Error:

• ❌ Model complexity

• ❌ Standard deviation

• ❌ Accuracy

• ✅ Average squared difference between predicted & actual values

---

4. Question 4

Primary goal of gradient descent:

• ❌ Standardize features

• ❌ Compute gradient of inputs

• ❌ Find maximum

• ✅ Minimize the cost function

---

5. Question 5

If learning rate is too large:

• ❌ Takes longer

• ❌ Converge suboptimally

• ✅ The algorithm may miss the minimum

• ❌ Converges too fast

Explanation:
Large LR → jumps over minimum → divergence.

---

6. Question 6

Why set requires_grad=True?

• ❌ For visualization

• ❌ Make immutable

• ❌ Improve performance

• ✅ To automatically compute gradients

---

7. Question 7

If learning rate is too small:

• ❌ Oscillations

• ❌ Large updates

• ❌ Rapid increases

• ✅ Convergence becomes very slow

---

8. Question 8

Cost surface represents:

• ❌ Data point plot

• ❌ Matrix

• ❌ Gradient

• ✅ Plot showing how parameters affect the cost function

---

9. Question 9

Role of the forward function:

• ❌ Initialize parameters

• ❌ Transform data

• ❌ Compute loss

• ✅ Compute predictions (forward pass)

---

10. Question 10

Significance of contour plots:

• ❌ Show data distribution

• ✅ Slices of the cost surface at different heights

• ❌ 3D view

• ❌ Visualize gradient directly

---

🧾 Summary Table

---