Problem
The playground tensorflow org website gives you a place to build and play around with simple neural networks. You can build networks with different types of data sets (see the pictures on the left under "DATA") and then design different kinds of neural networks.
For what's coming, I find it easier to check the "Discretize output" button over on the right under "OUTPUT."
Action Items
• For each of the 4 sets of test data, build a network with just two inputs (x1 and x2) and no hidden layers. About how many epochs does it take for the solution to converge (i.e., getting to a point where the solution isn't changing anymore)? Which data sets can it get right with no hidden layers?
• Recall that a network with no hidden layers is basically a perceptron, and a perceptron could only discover classifications that are linearly separable. How can you see that happening with these examples?
• For each of the data sets that was not correctly classified with no hidden layers, add hidden layers and other input parameters until you can get the input classified as best as you can. Specify what parameters you used. You might not be able to get them all perfect, but tell me what you did to get close. All things being equal, I'm looking for the simplest model that makes a correct classification.
The response should include a reference list. Using one-inch margins, Times New Roman 12 pnt font, double-space and APA style of writing and citations.