Machine Learning HW 2

Saving power is a fundamentally important goal in mobile devices where battery life limits the utility of such devices. Much research has been done in trying to develop schemes for saving power and extending battery life by turning off components of the device when they are not needed to save power. Choosing when to turn off the devices often requires a system to have the ability to predict when that device will be needed again. Spinning down a disk drive on a portable computer is an example of such a device which can benefit from schemes dedicated to predicting usage patterns. The authors of [1] introduce a type of machine learning algorithm called a share algorithm as it can be applied to the task of deciding when to spin down a disk drive. They have shown that such an algorithm can provide substantial decrease in energy consumption. In this paper, we have implemented the algorithm of [1] and explore its performance on the Cell01 data set. In addition, we have explored a variation that attempt to predict the mode of operation of the system to further improve the performance.