06 Prove : Assignment
Neural Network Part 1
Objective
Understand Neural Networks, in particular, the back-propagation algorithm for multi-layer perceptron networks.
Please note that you have 3 weeks to experiment with this algorithm. This is because it can be a challenge! This is likely the most challenging individual programming assignment we will do this semester. Do not put this off until later!
Instructions
You are welcome to use any off-the-shelf neural network library you like, but we strongly suggest that you use Keras, which is an API wrapper that makes using Google TensorFlow much easier.
Make progress toward your final experiments by completing the following this week:
Load at least one dataset of your choice. You should choose something more interesting than loading the built-in Iris dataset.
Appropriately prepare your dataset for a neural network, which may include handling missing data, normalizing numeric values, etc.
Also, make sure to split your dataset into a training and test set.
Identify a neural network library and learn about its parameters and how to use it.
Train a neural network with your dataset.
Use your network to make predictions about a test set and note the success of the algorithm on your dataset.
Experiment with at least 3 different sets of hyper-parameters for the algorithm (e.g., different number of layers and nodes per layer, different learning rates, different activation functions, etc.).
General Guidelines
Because we are not finishing the assignment at this point, this assignment does not specifically require (or reward) going above and beyond. That will be addressed in a future assignment.
You are, of course, encouraged to go beyond these expectations because that will help you finish the complete assignment.
Submission
When complete, answer the questions in the associated I-Learn quiz.
These questions simply ask you to enumerate which of the above steps you have completed.