Hands-On Python Deep Learning for the Web

A summary of the dataset

After a very concise exploration of the available dataset, we are able to come up with the following results.

The training dataset contains 60,000 images with a dimension of 60,000 x 784, where each image is 28 x 28 pixels. The distribution of samples among the digits are as follows:

Digit	Number of Samples	Digit	Number of Samples
0	5,923	5	5,421
1	6,742	6	5,918
2	5,958	7	6,265
3	6,131	8	5,851
4	5,842	9	5,949

 

Observe that digit 5 has a smaller number of samples than digit 1. So, it is quite possible that a model that isn't finely trained will make mistakes in recognizing digit 5.

The summary of the number of labels present tells us that all 60,000 ...