Processing numerical data

We will start by preparing numerical data. You have numerical data when:

• Your data is expressed by a floating number
• Your data is an integer and it has a certain number of unique values (otherwise if there are only few values in sequence, you are dealing with an ordinal variable, such as a ranking)
• Your integer data is not representing a class or label (otherwise you are dealing with a categorical variable)

When working with numerical data, a few situations may affect the performance of a DNN when processing such data:

• Missing data (NULL or NaN values, or even INF values) that will prevent your DNN from working at all
• Constant values that will make computations slower and interfere with the bias each neuron in the network is already providing
• Skewed distribution
• Non-standardized data, especially data with extreme values

Before feeding numerical data to your neural network, you have...