Companion code for Forward Thinking: Building and Training Neural Networks One Layer at a Time and Forward Thinking: Building Deep Random Forests submitted to NIPS 2017.
Authors:
We used a single desktop computer with:
run_mnist_cnn.sh
script.TODO: Make this pip installable.
Once the package has been installed, you may run the included run_mnist_cnn.sh
script.
This will run the forward thinking neural network (achieved 99.72% in our tests), the backpropagation equivalent of our model (achieved 99.63% in our tests), and saves and displays a plot comparing the test and train accuracies.