Initial Neural network with forward feeding
authorMartin Pitt <martin@piware.de>
Sat, 29 Aug 2020 10:48:59 +0000 (12:48 +0200)
committerMartin Pitt <martin@piware.de>
Sun, 30 Aug 2020 09:40:25 +0000 (11:40 +0200)
commit0ea12b213873b4bef12e1f2b65eed64704ee040f
treea100377304919a5ad749e67e8103c0b563788f39
parent8af4223121b60d5d67b7121d87c5c6fed01b58e7
Initial Neural network with forward feeding

Two hidden layers with parametrizable size. Two possible transfer
functions, defaulting to reLU for now.

Initialize weights and biases randomly. This gives totally random
classifications of course, but at least makes sure that the data
structures and computations work.

Also already add a function to recognize the test images and count
correct ones. Without trainingh, 10% of the samples are expected to be
right by pure chance.
README.md
train.py [new file with mode: 0755]