hox 1.2.0

Lightweight neual network library project.

HOX

HOX is not an alternative to big ml library like pytorch or tensorflow, it lacks features and optimization, such as gpu support. The goal is to create a lightweight library (< 100 lines of code) that is easy to use and quick to implement for creating small projects or experiment with ml.

pip install hox

---

examples/mnist

train.py

from hox import *
import utils

#Create model (2 layers, 784 input neurons, 144 first layer, 10 output layer)
model = Model.create([Layer(784, 144, Relu()), Layer(144, 10, Sigmoid())])

#Upload mnist dataset
X, Y, x, y = utils.mnist()

#Shuffle the dataset to improve training stability
indices = np.random.permutation(len(X))
X, Y = X[indices], Y[indices]

#Train the model
model.train(X, Y, epochs = 1, rate = 2, batch_size = 16)

#Save the trained model
model.save("mnist")

accuracy.py

from hox import *
import utils

#Load model
model = Model.load("mnist")

#Upload mnist dataset
X, Y, x, y = utils.mnist()

#Accuracy tested on test10k data (x, y)
counter = 0
for i in range(len(x)):
    if np.argmax(model.forward(x[i])) == y[i]:
        counter +=1
print(str((counter*100)/len(y)) + "% accuracy")