Predictive Modeling Using Neural Nets - Info and Demos
Did you miss Sam Gardner’s Feb. 10 live webcast on predictive modeling using neural nets? No problem. It’s available now as four short videos.
Part 1: Sam describes neural networks and characteristics of the JMP Neural platform.
Part 2: Sam demonstrates how to use single-layer neural nets to model device power output, including how to specify the validation method and the number of hidden nodes.
Part 3: Sam plots actual-by-predicted responses and residuals for the neural net, saves prediction formulas and more. He shows how to use validation and training sets, interpret confusion matrices and interpret confusion rates.
Part 4: Sam uses JMP Pro to develop complex, multi-layer neural networks. He uses linear and Gaussian activation functions, automates neural network sizing via boosting and normalizes/transforms irregular inputs.
Sam bundled his journal presentation, data tables and scripts as a JMP project. If you want to walk through his examples, simply download the .jmpprj file from the JMP File Exchange and open it in JMP. (SAS login is required to download.)
The Neural platform in JMP 9 and JMP Pro offers a rich set of modeling options, fast performance and calculations that streamline analysis. In a free white paper, JMP Statistical R&D Manager Chris Gotwalt describes the implementation algorithms and statistical methods used by the new JMP Neural platform. In the beginning of the paper, he identifies which options and capabilities are available only in JMP Pro.
Sam's presentation was one in a series of weekly Mastering JMP live webcasts. You can register for other Mastering JMP sessions that interest you. Visit my blog each Thursday to check for more information about the previous week's topic.