Simple test bench for perceptron NN,
port of python app. : https://python-course.eu/neural_network_mnist.php
I've tried to follow kiss principle for both perceptron (nn.nim) and linear algebra support (mat.nim) to make it educational, readable and adaptable, much cleaner and simple than .py
performance is remarkable (x4 python w/higher learning ratio) considering that
- no optimization or parallel resources have been used
- python implementation is basically a numpy wrapper
training and test data is included in a zip file, so:
unzip data.zip
before using
best performance has been obtained w/
nim c -r -d:release -d:danger nn
sources:
I agree with this.
Otherwise great job :)