Tai Li Hongyi--keras__ Li Hongyi

Keras Chinese Document
Keras English Document 1. Brief introduction

2. Keras Basic Flow

Take handwritten digit recognition as an example 1. Define Network structure

2. Set the form of loss function

3. Model Fitting

When batch_size=1, it is a random gradient descent stochastic gradient descent
We know that stochastic gradient descent is a lot faster than 50,000 data. However, when batch_size>1, it also increases the speed of the operation, for example, in handwritten digits recognition, the BATCH_SIZE=10 's final speed is greater than batch_size=1 speed. Why.
The reason is: GPU and matrix operation

If there is a GPU, then the larger the batch_size, the faster the operation. But the batch_size is too big for the effect to be particularly good. 4. Model Results