DEVELOPMENT OF R PACKAGE AND EXPERIMENTAL ANALYSIS ON PREDICTION OF THE CO2 COMPRESSIBILITY FACTOR USING GRADIENT DESCENT

Abstract

Nowadays, many variants of gradient descent (i.e., the methods included in machine learning for regression) have been proposed. Moreover, these algorithms have been widely used to deal with real-world problems. However, the implementations of these algorithms into a software library are few. Therefore, we focused on building a package written in R that includes eleven algorithms based on gradient descent, as follows: Mini-Batch Gradient Descent (MBGD), Stochastic Gradient Descent (SGD), Stochastic Average Gradient Descent (SAGD), Momentum Gradient Descent (MGD), Accelerated Gradient Descent (AGD), Adagrad, Adadelta, RMSprop and Adam. Additionally, experimental analysis on prediction of the CO2 compressibility factor were also conducted. The results show that the accuracy and computational cost are reasonable, which are 0.0085 and 0.142 second for the average of root mean square root and simulation time.

Keywords

• Gas compressibility factor
• Machine learning
• Regression
• R programming language