2. Model Evaluation

Model evaluation means checking how well a trained model performs on unseen data.

The model should be evaluated on data it did not train on.

Train vs Test

Common Metrics

Metric	Meaning
MSE	Average squared prediction error
MAE	Average absolute prediction error
Accuracy	Percentage of correct predictions
Precision	How many predicted positives were actually positive
Recall	How many actual positives were correctly found
F1 Score	Balance between Precision and Recall

Train and Evaluation Mode

During evaluation, disable training-specific behavior such as Dropout.

model.eval()

Disable gradient calculation because weights are not being updated.

with torch.no_grad():
    output = model(inputs)

Actual Implementation

import torch
import torch.nn as nn

class NeuralNetwork(nn.Module):
    def __init__(self):
        super().__init__()

        self.linear1 = nn.Linear(2, 10)
        self.dropout = nn.Dropout(0.2)
        self.linear2 = nn.Linear(10, 1)

    def forward(self, x):
        x = torch.relu(self.linear1(x))
        x = self.dropout(x)
        x = self.linear2(x)

        return x


model = NeuralNetwork()

inputs = torch.tensor([
    [18.0, 28.0],
    [19.0, 29.0],
    [20.0, 30.0],
    [21.0, 31.0]
])

targets = torch.tensor([
    [18.0],
    [19.0],
    [20.0],
    [21.0]
])

loss_function = nn.MSELoss()

# Evaluation mode
model.eval()

# No gradients during evaluation
with torch.no_grad():

    predictions = model(inputs)

    test_loss = loss_function(
        predictions,
        targets
    )

print("Test Loss:", test_loss.item())

Quick Difference

Training   → Update model weights
Evaluation → Measure model performance
model.train() → Training mode
model.eval()  → Evaluation mode
no_grad()     → Don't calculate gradients