stacked-autoencoders

Contains scikit-learn wrappers for StackedDenoisingAutoencoders and DenoisingAutoencoder.

StackedDenoisingAutoencoders are trained greedily layer by layer.

See example usage by running python run_mnist_ae.py which trains a single layer autoencoder on the MNIST dataset and uses the latent features as input to a naive bayes classifier.

Similarly, run python run_mnist_stacked_ae.py for a stacked two layer autoencoder.

Tested with:

• Python 2.7.9
• Theano 0.7.0
• Scikitlearn 0.17
• Numpy 1.10.2