Chapter 1 Introduction

In this tutorial you will be introduced to several R packages and how to use some of them such as neuralnet , keras and h2o .

In any of these packages the order of NN application is standard:

1 - Preprocess your data:

  • Standardize or normalize your data (scale or min-max)

  • Missing values, outliers

  • Training/Validation and Test set splits

2 - Construct your NN model, ie. number of neurons, number of layers, learning rate, regularization etc. (these are your hyperparameters)

3 - Fit your model

4 - Predict your Y variable for both training and validation sets.

5 - Extract the training and validation errors, visualize these for different number of hyperparameters (fine tuning)

6 - Choose the parameters based on the smallest validation error.

7 - Assess the performance of your predicted model on the test set. Remember the test set error is not for decision making!