https://github.com/jaanli/vimco_tf
VIMCO in tensorflow.
https://github.com/jaanli/vimco_tf
machine-learning mnist tensorflow
Last synced: 8 months ago
JSON representation
VIMCO in tensorflow.
- Host: GitHub
- URL: https://github.com/jaanli/vimco_tf
- Owner: jaanli
- Created: 2016-11-04T19:26:12.000Z (about 9 years ago)
- Default Branch: master
- Last Pushed: 2016-12-02T21:28:48.000Z (about 9 years ago)
- Last Synced: 2025-04-12T02:06:11.016Z (8 months ago)
- Topics: machine-learning, mnist, tensorflow
- Language: Python
- Size: 31.3 KB
- Stars: 5
- Watchers: 2
- Forks: 2
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# vimco_tf
Variational Inference for Monte Carlo Objective (VIMCO) in tensorflow. The paper is here: https://arxiv.org/abs/1602.06725
This gets to a log-likelihood of `-94.3` nats on the validation set of the binarized MNIST data.
## How to run
Important: needs to be run with a tensorflow version of [at least 0.11.0](https://www.tensorflow.org/versions/r0.11/get_started/os_setup.html#pip-installation).
```
# get the binarized MNIST dataset, save to /tmp/binarized_mnist.hdf5
python make_binarized_mnist_hdf5_file.py
# run sbn training with vimco. ideally on GPU (10x speedup)
python sbn_vimco.py
# visualize logs
tensorboard --logdir /tmp
```
Summaries and posterior predictives can be viewed on tensorboard:
This is heavily based off of Joost's implementation at https://github.com/y0ast/VIMCO (thank you Joost!)