RuiShu / bcde

Bottleneck Conditional Density Estimation (ICML 2017)

bcde

The Bottleneck Conditional Density Estimator provides a semi-supervised learning framework for high-dimensional conditional density estimation. This repository provides code to run experiments in from the 2017 ICML paper Bottleneck Conditional Density Estimation.

This work was done while interning at Adobe Systems.

Dependencies

Please make sure to pip install the following dependencies

tensorflow-gpu==1.1.0
tensorbayes==0.1.1

Example

To run the model (2-layer hybrid+factored BCDE), simply do:

python main.py --model hybrid_factored