GradDescentR: An R package implementing gradient descent and its variants for regression tasks

Abstract

Gradient Descent (GD) is one of famous machine-learning methods used for prediction on regression tasks in many fields. However, only a few software library utilizing it can be found in the literature. Therefore, this research is aimed to implement the method and the following variants: Mini-Batch Gradient Descent (MBGD), Stochastic Gradient Descent (SGD), and Stochastic Average Gradient Descent (SAG). The package is written in the programming language R, which is a software environment offering many facilities for statistics and machine-learning tools. Moreover, we provide the use of these approaches in a case study of gas industries, which is to predict values of the compressibility factor (Z-factor) of CO 2 . Basically, we consider the problem as a regression task including pressure and temperature as the input attributes. The first step is generating a learning model from available training data by using the methods. Then, the model is used to predict the Z-factor over new data. A performance comparison between the package and other methods from the literature is presented as well.