Neural network to classify MNIST handwritten digits using machine learning in Python 3. Includes convolutional and fully-connected sigmoid layers. (intended as an educational project).
Requires Numpy, a linear algebra library, and the MNIST dataset.
main.py: Train and evaluate network on MNIST dataset using stochastic gradient descent.
-
input: set of 28x28 input images x and expected outputs y.
-
output: 10x1 vector of probabilities (activations) of the input image x corresponding to each digit from 1 to 10. This is the network's guess, whereas y is the correct answer.
-
basic_neuralnet.py: basic network implementation with fully-connected neurons. -
neuralnet.py: modular neural network implementation with convolutional and fully-connected sigmoid layers.