Modeling Power in Batteries with Linear Regressions

A fundamental aspect of how Deep Learning systems work is that they make predictions about how a function should fit some test dataset, evaluate the errors from that function to the test dataset, and modify that function to reduce the errors. In science, we also do this when thinking about the best ways to design an experiment, or to probe relationships between different variables. In this lesson, students will vary the number of batteries in an electrical circuit, and compare the expected amount of power to the actual electrical power. By varying the type of battery, students can gain a familiarity with how input power does not equal output power. Next, they will use a linear regression to interpolate and extrapolate power outputs of batteries. Finally, they will also explore the “bias” inherent in their model (that they used data about AAA batteries, and so it might not be a very good predictor for 9V batteries), and relate that understanding to how deep Machine Learning and Deep Learning algorithms function and are affected by their data sources.