driver.py

import sys

sys.path.append("./src/")
from neuralnet import NeuralNetwork
from src.book.mnist_loader import load_data_wrapper
from train import test_evaluation, train

train_data, valid_data, test_data = load_data_wrapper()

# loss_fn = mse, bce, ce (ce and softmax should be used together)
# weight_init = he, xavier, uniform
network = NeuralNetwork(
    network_shape=(784, 128, 10), lr=0.1, loss_fn="ce", weight_init="he"
)


# final_layer_act = sigmoid, relu, tanh, softmax
# intermediate_act = sigmoid, relu, tanh
# activations = (intermediate_acts, final_acts) | activations = (one_act_for_all, )
network.activation_setup(activations=("relu", "softmax"))
# regularization = l1, l2 | reglambda = (0.0-1.0)
# droput = (0.0-1.0) (percent of nodes to dropout in each layer)
network.regularization_setup(regL="None", reglambda=1e-5, dropout=0.0)
# optimizer_name=name, args=(args for that optimizer)
# sgd_momentum args = (momentum = 0.0 - 1.0)
# adam args = (*beta1* (for first moment [mean]) = 0.0-1.0,  *beta2* (for second moment [variance]) = 0.0-1.0)
network.optimizer_setup(optimizer_name="sgd_momentum", args=(0.9,))
# network.optimizer_setup(optimizer_name="adam", args=(0.9, 0.999))

train(network, train_data, epochs=2, batch_size=8)

print("Train Accuracy: ", test_evaluation(network, train_data), "%")
print("Validation Accuracy: ", test_evaluation(network, valid_data), "%")
print("Test Accuracy: ", test_evaluation(network, test_data), "%")