Create Readme

[Inominie]

Create a readme file to lay out setup procedures and intuitions of the project

[josephcoveai]

I first tried a deeper neural network with lots of nodes and experimented with different activation functions. I found that ReLU seemed to be the most reliable. I also started by using a Dropout of 0.5. I found that what had the most positive effect on my results was the convolutional layers. So, I did two convolutional layers with a pooling layer in between them. I lowered my dropout to 0.1 and placed it in between the input and output layers in my neural network. I found that smaller neural networks can actually be more effective than larger ones, which seemed counter-intuitive.