It is a very simple neural network with two hidden layers. The advantage of using mid level API is that it gives the flexibility to modify layers and at the same time, avoids the complication of defining each layers (weights, biases etc.) manually.
MNIST Fashion data set has been used.
- Tensorflow >= 1.6
- Numpy
- sklearn
A portion of code is taken from http://miguelromao.me/2018/03/20/convolutional-neural-networks-for-the-mnist-dataset-tensorflowlayers-and-keras/ and thanks to sentdex for the detailed explanation.