1/19/17

Today I finished the first neural network project in my course. I got to see the neural network "in action" in a sense; I watched as the neural network repeatedly trained itself on the training data. The network trains itself in what are called "epochs"; one epoch consists of the network training itself on the entire training data set one time. The neural network I was working with went through 100 epochs of training. Each epoch marginally improves the network's accuracy in its predictions. As I mentioned in previous blogs, this network's purpose was to predict whether or not a customer would leave a bank based on their credit score, location, gender, tenure, salary, etc. After finishing this neural network, I tested it on the "test" data set (a subset of the overall data set), and it predicted whether or not a certain customer would leave a bank with around an 85% accuracy, a very impressive result. After testing the network on the test data set, I tested it on a specific data point not included in the data set to see what prediction it would make. Next class, (after the class periods where we present), I will be completing the next step: fine-tuning the network and evaluating its performance.