A simple 3-layer ANN (artificial neural network) written in Python.
Note: I have written this same 3-layer neural network in Go which you can find here.
- python 2.7 (I haven't tested any other version)
- numpy
- scipy
The example uses the MNIST database to train and test the neural network.
The MNIST (Modified National Institute of Standards and Technology) database contains 60,000 training images and 10,000 testing images of handwritten numbers from 0-9.
Starting neural network to recognize handwritten digits.
Training the neural network.
Training epoch 1/1.
complete.
Testing the neural network.
complete.
Neural network is 95.94% accurate at predicting handwritten digits.
The example takes ~45s to complete.
Note: the same setup in Go can be found here and it takes ~85s. The numpy library has been so optimized when dealing with complex mathematics that it is hard to do better, even in a compiled language, even when coding for particular use cases.