Exercise

🐧 Train a Small Neural Network on the Penguins Dataset

Goal: Build and train a simple neural network (using PyTorch) to classify penguin species based on their physical measurements. You will then evaluate how well your model performs.

1. Data Preparation:

• Load the dataset (e.g., using seaborn)
• Drop rows with missing values
• Select numerical features such as:
  • bill_length_mm
  • bill_depth_mm
  • flipper_length_mm
  • body_mass_g
• Encode categorical variables (e.g. sex)
• Split into train and test sets

2. Build a Small Neural Network (PyTorch)

Create a tiny feed‑forward network, for example:

• Input: 4 features
• Hidden layer: 16 neurons + ReLU
• Output: 3 classes (Adelie, Gentoo, Chinstrap)

Example skeleton:

import torch
import torch.nn as nn

class PenguinNet(nn.Module):
    def __init__(self):
        super().__init__()
        self.model = nn.Sequential(
            nn.Linear(4, 16),
            nn.ReLU(),
            nn.Linear(16, 3)
        )

    def forward(self, x):
        return self.model(x)

3. Train the Model

• Use CrossEntropyLoss
• Use Adam optimizer
• Train for 200–300 epochs

Track the loss over time and plot the training loss curve.

4. Evaluate the Model

On the test set:

• Compute accuracy
• Generate a confusion matrix (use sklearn.metrics.confusion_matrix)
• Interpret the results:
  • Which species are easiest to classify?
  • Which ones get confused with each other?