predict which digit is from the images Keras Tensorflow was used as the deeplearning framework Did same model building only using Tensorflow without Keras
Pixels of each image,
Convolution neural network Activation relu, added two dense layers with size of of the x_train index so number of pixels, dropped half first and then 1/3 added a extra layer of 10 with activation softmax loss was categorical crossentropy optimizer adam and metrics was accuracy.