I am a novice, may I ask when training, the bigger the batch_size is better or better 1

[clxa]

I am a novice, may I ask when training, the bigger the batch_size is better or better 1

[pranavbudhwant]

Let's consider the extreme scenarios:

1. The largest batch size, which is equal to the size of your training data - the gradient updates will be slower but less noisy.
2. The smallest batch size, which is equal to 1 - the gradient updates will be quick, but it'll be noisier.

So ideally you want to find a batch size where the updates aren't that noisy but are also fast.

[clxa]

让我们考虑一下极端情况：

1. 最大批次大小，等于训练数据的大小-梯度更新将较慢，但噪声较小。
2. 最小批量大小，等于1-梯度更新将很快，但是会更嘈杂。

因此，理想情况下，您希望找到一个批量大小，使更新不那么嘈杂但又很快。

Thank you for your reply. If my trainB is all of the same style images, can I ignore the batch -- size