Backprop

ACML Assignment

Requirements:

• Python
• Pyplot ( if visualization of the weights is wanted )

For this assignment we are tasked to write our own augmented neural network using the knowledge about neuron activation, the sigmoid function, and forward- and back-propagation.

Run instructions

To run this program, all you need to do is run the main method of main.py.

There are two parameters in the main method to play around with:

• rates: this is the learning rate of the neural network
• iterations: this is how often the network will iterate over the training set
• decays: this are the weight decays

When running the program the user is prompted to choose whether to run a GridSearch. This GridSearch will try multiple options for the parameters mentioned above, or when the user declines using the GridSearch, a single run is done with the optimal parameters that are (hard)coded in the main method of the program.

After training

The program prints the final error rate of all parameters and produces a plot of the error rates throughout all iterations.

If a single run of the program was chosen by the user, the network and it's weights will be visualized.