Epochs

What Is an Epoch?

An epoch means that the neural network has gone through the entire training dataset once.

For example, suppose we have 5 training examples:

Training Data

Example 1
Example 2
Example 3
Example 4
Example 5

When the neural network processes all 5 examples once, that is 1 epoch.

Example 1
Example 2
Example 3
Example 4
Example 5

        ↓

1 Complete Pass

        ↓

1 Epoch

A Simple Example

Imagine that we want to train a neural network to predict whether a student will pass based on study hours.

Student    Study Hours    Result

A          1              Fail
B          2              Fail
C          4              Pass
D          5              Pass
E          7              Pass

We have 5 training examples.

If the neural network processes all 5 students once, it has completed one epoch.

What Happens During One Epoch?

During an epoch, the neural network goes through the training data and learns from it.

Training Data
      ↓
Student A
      ↓
Prediction → Loss → Weight Update

Student B
      ↓
Prediction → Loss → Weight Update

Student C
      ↓
Prediction → Loss → Weight Update

Student D
      ↓
Prediction → Loss → Weight Update

Student E
      ↓
Prediction → Loss → Weight Update

      ↓

Epoch 1 Complete

After the network has processed the entire dataset once, the first epoch is finished.

Why Do We Need Multiple Epochs?

One pass through the training data may not be enough for the neural network to learn the patterns properly.

So we allow it to see the same training data again.

Epoch 1
  ↓
Network learns a little

Epoch 2
  ↓
Network improves

Epoch 3
  ↓
Network improves more

Epoch 4
  ↓
Network improves again

Epoch 5
  ↓
Network may become much better

Each epoch gives the model another opportunity to adjust its weights and reduce its errors.

Epochs and Loss

During training, we often watch the loss to see whether the model is improving.

Epoch 1
Loss = 1.20

Epoch 2
Loss = 0.80

Epoch 3
Loss = 0.52

Epoch 4
Loss = 0.31

Epoch 5
Loss = 0.20

In this example, the loss is decreasing as training continues.

That suggests the model is learning from the training data.

Easy Way to Understand an Epoch

Think about studying a textbook.

Read the entire textbook once
        ↓
1st pass

Read the entire textbook again
        ↓
2nd pass

Read the entire textbook again
        ↓
3rd pass

Each complete pass through the textbook is similar to an epoch.

Similarly, each complete pass through the training dataset is one epoch.

Dataset vs Epoch

These two terms are different.

Dataset
↓
The complete collection of training examples


Epoch
↓
One complete pass through that dataset

For example:

Dataset = 1,000 training examples

1 Epoch
= Process all 1,000 examples once

5 Epochs
= Process the 1,000 examples five times

Epochs in Python

In Python, we can represent multiple epochs using a loop.

epochs = 5

for epoch in range(epochs):

    print("Epoch:", epoch + 1)

This runs the training process five times.

Epoch: 1
Epoch: 2
Epoch: 3
Epoch: 4
Epoch: 5

In a real neural network, the training code would run inside this loop.

Epochs Inside a Training Loop

A simplified training loop can look like this:

epochs = 5

for epoch in range(epochs):

    prediction = model(training_data)

    loss = calculate_loss(prediction, target)

    gradients = calculate_gradients(loss)

    update_weights(gradients)

    print("Epoch:", epoch + 1)
    print("Loss:", loss)

The important part is that the complete training process is repeated for each epoch.

Epoch
  ↓
Prediction
  ↓
Loss
  ↓
Gradients
  ↓
Weight Update
  ↓
Next Epoch

What If We Use Too Few Epochs?

If we stop training too early, the neural network may not have learned enough from the training data.

Epoch 1
Epoch 2

        ↓

Training stops too early

        ↓

Model may still have high loss

        ↓

Model may not have learned enough

This situation is commonly associated with underfitting.

Can We Use Too Many Epochs?

Yes. More epochs are not automatically better.

If we keep training for too long, the model may start fitting the training data too closely and perform worse on new, unseen data.

Too Few Epochs
      ↓
Model has not learned enough


Good Number of Epochs
      ↓
Model learns useful patterns


Too Many Epochs
      ↓
Possible overfitting

This is why we do not simply choose the largest possible number of epochs.

Important: An Epoch Is Not One Weight Update

This is a common beginner mistake.

One epoch means one complete pass through the dataset. The number of weight updates depends on how the data is divided into batches.

Epoch
  ↓
Many training examples
  ↓
Processed in batches
  ↓
Weight updates
  ↓
Complete dataset processed
  ↓
Epoch complete

We will learn exactly how this works when we study batch size and iterations.

Epoch vs Iteration

Do not confuse these terms.

Epoch
→ One complete pass through the entire dataset.


Iteration
→ One training step using one batch of data.

For example, if we have 100 training examples and use batches of 20:

100 examples
÷
20 examples per batch
=
5 iterations per epoch

So:

1 Epoch
= 5 Iterations

10 Epochs
= 50 Iterations

We will study this relationship in detail in the next topics.

Remember This

Epoch
=
One complete pass through the entire training dataset.


Example:

1,000 training examples

1 Epoch
= Model processes all 1,000 examples once.

10 Epochs
= Model processes all 1,000 examples ten times.

The simplest definition to remember is: one epoch = one complete pass through the training dataset.