Auto Encoder on Tensorflow
This package provides:
- Basic AutoEncoder
- Denoising AutoEncoder
- Stacked (Denoising) AutoEncoder
#Requirements
- Python 2.7 or 3.3+
- TensorFlow >= 0.6.0
#Installation
Clone git repository and run setup.py as the following commands.
$git clone https://github.com/nukui-s/tfautoencoder.git
$cd tfautoencoder
$sudo python setup.py install
#Basic Usage
import numpy as np
from tfautoencoder import TFAutoEncoder
from tfautoencoder import TFStackedAutoencoder
#prepare data as numpy array or pandas DataFrame
data = np.random.rand(1000,50)
#If you want to encode 50 dim into 10 dimension
ae = TFAutoEncoder(hidden_dim=10)
#fitting weight and bias in AutoEncoder
ae.fit(data)
#encode data
encoded = ae.encode(data)
#reconstuct data
data2 = ae.reconstruct(data)
#define Denoising AutoEncoder
dae = TFAutoEncoder(hidden_dim=10, noising=True)
#Stacked AutoEncoder with 100, 50, 30 layer units
sae = TFStackedAutoEncoder(layer_units=[100, 50, 30])