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

---

EXAMPLE

You can easly test the library on the MNIST example with the code found at examples/mnist

• pip install hox to install the module
• python .\train.py to create and train the nerual network
• python .\accuracy.py to test the model accuracy on the test dataset

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")