What Is Epoch, Batch Size and Batch?

原创于 2019-03-21 21:42:10 发布 · 222 阅读

0 ·

CC 4.0 BY-SA版权

深度学习专栏收录该内容

2 篇文章

订阅专栏

博客介绍了神经网络训练中使用Epochs和Batch Size的原因。因训练数据集大，需将其分成小份以更新神经网络权重。还给出了Epochs、Batch Size和Batch的定义，并举例说明，如2000个样本的数据集，以500为Batch Size，需4次迭代完成1个Epoch。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

1. Why Do We Use Epochs and Batch Size?

Training dataset sometimes may be very large for the computers to deal with, so we have to divide the dataset into smaller size to help the computers update the weights of the neural networks each time. That is why we use epochs and batch size.

2. Epochs

Definition: One Epoch is when an ENTIRE dataset is passed forward and backward through the neural network only ONCE.

The number or Epoch gives no clear conclusions because datasets and the diversity of them are different.

3. Batch Size

Definition: Total number of training examples present in a single batch.

NOTE: Batch Size and Number or batches are two different definitions.

4. Batch

Definition: Batch is the number that you devide your all data into several parts.

5. Example

We can divide the dataset of 2000 examples into batches of 500 then it will take 4 iterations to complete 1 epoch.

Batch Size: 500

Itertaion: 4