You are on page 1of 4

Preprocessing of mnist dataset

• Mnist dataset(images) are already centered and they all have the
same size(No scaling and centering needed).

• But the default dataset is somewhat inconveniently formatted.so


we have created a csv-formatted version.

Mnist in csv
Preprocessing of dataset
• Image files can not be easily manipulated.

• We have converted the images into 2d matrix represented as their


pixel values.

After converting
Understanding the dataset
Size of image: 28 X 28 pixel(total 784 features)

label : This is actual digit number this


handwritten digit image represents. It is
either 0 to 9.
Understanding the dataset
• Dataset are divided randomly into two subset i.e
training and test set.

• Dataset have total 70000 data, where training


dataset size is 60000, and test dataset size is 10000.
Pixel values/data labels
70000 values

You might also like