NNetOneSplit

Python3 version 3.6.1 or greater is REQUIRED

Our NNetOneSplit function is located at: https://github.com/jguzman-tech/sgd-neural-network/blob/master/main.py

1. Environment setup

These steps assume a linux system.

1.1. First clone the repo

git clone https://github.com/jguzman-tech/sgd-neural-network.git

1.2. Create an arbitrary directory

mkdir ./.venv

1.3. Create a virtual environment

python3 -m venv ./.venv

1.4. Activate the virtual environment

. ./.venv/bin/activate

1.4.1 You can use an easy to remember bash alias to activate the venv

1.4.1.1 Open your ~/.bashrc

vim ~/.bashrc

1.4.1.2 Add this line to your ~/.bashrc

alias act=". ./.venv/bin/activate"

You can replace act with any valid bash identifier you want.

1.4.1.3 Now reload your ~/.bashrc

source ~/.bashrc

1.4.1.4 Now you can use the “act” command

act

This will be saved on all future shells sessions for your current user.

1.5. Install module requirements

pip3 install -r ./requirements.txt

2. Execution

To see our help message execute: python3 main.py -h

prompt$ python3 main.py -h
usage: main.py [-h] [--use-custom-ll] max_epochs step_size n_hidden_units seed

Use the SGD algorithm on the spam.data set

positional arguments:
  max_epochs       The maximum number of epochs
  step_size        The scaling factor used for adjusting weights
  n_hidden_units   The number of hidden parameters in our hidden layer
  seed             The seed used for our random number generator

optional arguments:
  -h, --help       show this help message and exit
  --use-custom-ll  Set this flag if you want to calculate using the LL
                   function we coded.We used the library version to prevent
                   overflow.

3. Reproduce our results

python3 main.py 500 0.05 10 4

The resultant figure will be: epochs_500_step_0.05_units_10_seed_4_logistic_loss.png

As you can see the filename includes all of the arguments so you can easily identify them.