xemptuous / zig-ann

An example of using Zig to create a simple ANN

zig-ann

An example of using Zig to create a simple ANN with 1 hidden layer simulating an XOR gate.

Activation Function - Hyperbolic Tangent (tanh)

Loss Function - Sum of Squared Errors (SSE)

Adjusting

Global variables at the top of main.zig can be used to tune the ANN.

BIAS : The global bias for each neuron.

ALPHA : The "learning rate"; the smaller the number, the more accurate the results, but the longer it takes to train.

NUM_HIDDEN : The number of hidden neurons in the hidden layer.

NUM_INPUTS : The number of inputs fed to the neurons in the hidden layer.

NUM_OUTPUTS : The expected number of outputs in the output layer.

NUM_TEST_CASES : The number of test cases being fed to the model to be used in each epoch.

ERROR_THRESHOLD : The threshold of the loss function that signals a successful training and an exiting of the function.

EPOCH_THRESHOLD : The number of epochs before restarting and rerolling of weights occurs.

EPOCH_PRINT_THRESHOLD : The number of epochs that must pass before printing status updates.