Untapped ======= Untapped provides a sklearn-compatible class for constructing a semi^2-supervised deep generative model built with neural networks and trained with variational inference.

Untapped depends heavily on the Lasagne and Theano libraries. Please make sure you have these installed before installing Untapped. Untapped also depends heavily on a modifed version of Parmesan. We recommend installing Untapped in a virtual environment (i.e., download Anaconda).

Install Untapped

git clone https://github.com/all-umass/untapped.git
cd untapped
conda create --name untapped python
source activate untapped

pip install -r requirements.txt
python setup.py install

OR

pip install -r requirements-dev.txt
python setup.py develop
python setup.py test

At the moment Untapped primarily includes

• A semi^2-supervised deep generative model (^2 for unlabeled x and unlabeled y) equipped with training procedures for both variational inference and vanilla reconstruction error. Models M1, M2, and M12 from Kingma et. al are implemented.

Please see the source code and code examples for further details.

• examples/DEMO_crism.py: M2 model trained on hyperspectral data obtained under CRISM (Mars satellite instrument) like conditions
• examples/DEMO_libs.py: M2 model trained on LIBS spectral data obtained from Mars Curiosity rover
• examples/DEMO_raman.py: M2 model trained on Raman spectral data of mineral mixtures
• examples/DEMO_mnist.py: M2 model trained on MNIST data with half labels missing (never see 5-9)

Usage example: python examples/DEMO_crism.py

Untapped is a work in progress, inputs, contributions and bug reports are very welcome.

The library is developed by

• Ian Gemp

• Kingma, D. P., & Welling, M. (2013). Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114.
• Burda, Y., Grosse, R., & Salakhutdinov, R. (2015). Importance Weighted Autoencoders. arXiv preprint arXiv:1509.00519.
• Rezende, D. J., & Mohamed, S. (2015). Variational Inference with Normalizing Flows. arXiv preprint arXiv:1505.05770.