Three implementations of perceptrons:
-
The Logical Perceptron creates a logical operator learning perceptron, i.e. it learns how to compute "and", "or", "nand", and "nor". It returns its weights and test sample performance so that training can be seen quantitatively. For more understanding, it also plots the movement of the decision boundary that linearly separates the data.
-
The Cluster Perceptron seperates two clusters of data. It also prints out the weights and the results of the training on both the training and test sets. In theory, one can input their own cluster data and it will work out of the box.
-
The Multilayer Perceptron, better known as a Neural Network, can be built to any desired network size. The activation functions, weight initializations, and learning rates are customizable at the layer level. As always, graphs will be included.
Below are pictures showing prototypical output from the algorithms:
