Nural

Nural is a small Deep learning framework I built for fast Neural Network creation and evaluation. This framework is built around numpy in order to aid fast CPU compute

The design of this framework allows easy notations, allowing the user to create, train an evaluate models with ease.

from nuralengine import Layer
import nuralengine
import numpy

model = nuralengine.Network([
    Layer.Input(40),
    Layer.Dense(42,activation=nuralengine.sigmoid),
    Layer.Dense(100),
    Layer.Dense(20),
    Layer.Dense(2)])

model[numpy.random.randn(40)] = numpy.ones(2)
# equivalent square(model(numpy.random.randn(40)) - numpy.ones(40)).mean().backward().step()

Nural also has a simple and elegant notation for Adverserial Networks

with Adverserial(gen):
    # G(X) must be tuned for D(G(X)) == 1
    gen[np.random.uniform(-1,1,5)] &= des,1

des[gen[np.random.uniform(-1, 1, 5)]] = 0
    

Give it a try by running main.py

python main.py

Contributions

This project is very much open contributions. Feel free to report bugs.