How it works...

The key steps that we have performed in the preceding code are as follows:

• We flattened the input dataset so that each pixel is considered a variable using the reshape method
• Additionally, we scaled the dataset so that each variable now has a value between zero and one
  • We achieved the preceding by dividing the values of a variable with the maximum value of that variable

• We performed one-hot encoding on the output values so that we can distinguish between different labels using the to_categorical method in the np_utils package
• We built a neural network with a hidden layer using the sequential addition of layers
• We compiled the neural network to minimize categorical cross entropy loss (as the output has 10 different categories) using the model.compile method
• We fitted the model with training data using the model.fit method
• We extracted the training and test losses accuracies across all the epochs that were stored in the history
• We also identified a scenario that we consider overfitting