This was written as ANN learning experience project, and is by no means a super efficient ANN module (yet). It currently can implement a perceptron model or a ANN with a single hidden layer.
Use the weights file weights_perceptron.npy
if using the perceptron network.
Use the weights file weights_multilayer.npy
if using the multilayer network.
Requires the numpy module and python 3.6 or newer. This can be installed with pip:
> pip install numpy
If you don't have pip, here are details on installing numpy: https://scipy.org/install.html
usage: pyann.py [-h] [--weights WEIGHTS]
{train,test} {perceptron,multi-layer}
positional arguments:
{train,test} command to run
{perceptron,multi-layer}
what type of ANN to run
optional arguments:
-h, --help show this help message and exit
--weights WEIGHTS load a weights file instead of rng weights
note: Some settings such as learning rate, train/test file names are located at the top of project2.py as constants.
Data files must be tab delimited and in the format <latitude> <longitude> <class>