MicroMLP is a tiny, lightweight python framework for creating and training feed-forward neural networks. MicroMLP works with custom activation and loss functions, but comes with the most common already provided. There is no support for hardware acceleration, but micromlp makes use of numpy for efficient matrix operations.
MicroMLP provides a simple API inspired by the Keras Sequential API found in Tensorflow.
import micromlp
from micromlp.activations import relu, softmax
model = MLP([
Layer(2, 8, relu),
Layer(8, 8, relu),
Layer(8, 2, softmax)
])And training the model is as simple as a single line of code.
from micromlp.losses import cross_entropy
model.train(xs, ys, loss=cross_entropy, epochs=20, batch_size=32, learning_rate=0.01)MIT