A minimal feedforward neural network from scratch using NumPy.
- Activations: sigmoid, tanh, ReLU, softmax, identity
- Training: mini-batch SGD with backpropagation
- Classification & regression support
- Save/load via pickle
from nn import Net, Activation
net = Net()
net.build(layers=[784, 128, 64, 10],
activations=[Activation.relu, Activation.relu, Activation.softmax])
net.train(data, labels, rate=0.01, epochs=50, batch_size=32, classification=True)
net.save("model.pkl")
net.load("model.pkl")