First, here are the required imports:
import tensorflow as tfimport numpy as np
Next, we acquire and preprocess the data. Notice that fashion_mnist is handily present in tf.keras.datasets. The x values in the dataset are in the form of integer NumPy arrays, each element being in the range 0 to 255, representing the grayscale value at each of the pixels in the 28 x 28 pixels fashion image. For training purposes, these values must be converted into floats in the range of 0 to 1. The y values are in the form of unsigned 8-bit integers (uint8) and must be converted into 32-bit integers (int32), again for use by the Estimator.
The learning rate is set to quite a small value, although this hyperparameter value may be experimented with: ...
