README.md

Loulou

Getting started

Description :

This python projects aims to bring a new deep learning implementation to frond-end developers.
It currently supports MNIST database and then can train neural networks on it, and predict grayscale 28*28px image that represents a number from 0 to 9.

Prerequisites :

You must install numpy to use Loulou.
In intend to show image that is predicted, you must also install matplotlib.

From pip :

pip install numpy
pip install matplotlib #(optional)

Installation :

To install Loulou, just do :

git clone https://github.com/aunetx/loulou
cd loulou/

Utilisation :

There are two ways to use loulou :

Training a model :

To train a model, just do :

python ./scripts/train.py -e [number of epochs] -b [batch size] -l [learning rate] -a [architecture] -f [filename.npy]

Or, for windows :

py .\scripts\train.py -e [number of epochs] -b [batch size] -l [learning rate] -a [architecture] -f [filename.npy]

architecture is a list of digits that defines the number of neurons in each hidden layer.

For example : python ./scripts/train.py -a 400 200 100 creates a network with 5 layers like :

1. 784 neurons - Input layer
2. 400 neurons - First hidden layer
3. 200 neurons - Second hidden layer
4. 100 neurons - Third hidden layer
5. 10 neurons - Output layer

All the arguments are optional : if you don't set a filename, the training is not saved.

Making a prediction :

This is as simple as :

python ./scripts/run.py [path/to/weights/file.npy] [path/to/image.png]

Or, for windows :

py .\scripts\run.py [path\to\weights\file.npy] [path\to\image.png]

To go further :

Version :

• Version 1.1.2 : output improvement for training added