Python AutoEncoder from scratch using Numpy

Usage

1. Install requirements :

$ pip install -r requirements.txt

2. Optionally run mnist_test.py.

$ py mnist_test.py 

Training

Instatiate an Autoencoder object :

from autoencoder import Autoencoder

autoencoder = Autoencoder(in_len=300, bottleneck=50, 0.001, relu)

And then via the train_dataset method to train over a dataset :

autoencoder.train_dataset(data)

Or via the train to input each data points iteratively :

autoencoder.train(v)

Inference

Use your Autoencoder object with the encode and decode methods like so :

example = ...
code = autoencoder.encode(example)
output = autoencoder.decode(code)