# 12/2017: Tensorflow distributions

**URL:** <https://discourse.pymc.io/t/12-2017-tensorflow-distributions/769>\
**Category:** Journal Club\
**Created:** [January 21, 2018, 2:18pm UTC](https://discourse.pymc.io/t/12-2017-tensorflow-distributions/769 "2018-01-21T14:18:13Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![junpenglao](https://yyz2.discourse-cdn.com/flex036/user_avatar/discourse.pymc.io/junpenglao/32/8_2.png) [@junpenglao](https://discourse.pymc.io/u/junpenglao)\
**Post date:** [January 21, 2018, 2:18pm UTC](https://discourse.pymc.io/t/12-2017-tensorflow-distributions/769/1 "2018-01-21T14:18:13Z")

</div>

Collection of notes and related information on the discussed paper: [TensorFlow Distributions](https://arxiv.org/abs/1711.10604)

1. The [bijector class](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/ops/distributions/bijector_impl.py) of tfb is quite a powerful concept. For example, two recent blog post [[1]](http://blog.evjang.com/2018/01/nf1.html) [[2]](http://blog.evjang.com/2018/01/nf2.html) from [Eric Jang](https://twitter.com/ericjang11/status/953774252527779840) demonstrated how to use it to construct Flows to get expressive distribution.

2. The implementation of [HMC in tensorflow](https://github.com/tensorflow/tensorflow/blob/r1.4/tensorflow/contrib/bayesflow/python/ops/hmc_impl.py), the leapfrog step is done inside of Tensorflow (as a comparison, PyMC3 also has a version to [perform leapfrog integration inside the computation graph](https://github.com/pymc-devs/pymc3/blob/master/pymc3/step_methods/hmc/trajectory.py#L12), however, the current implementation performs [leapfrog outside of theano](https://github.com/pymc-devs/pymc3/blob/master/pymc3/step_methods/hmc/integration.py)). See alos the [related discussion on Github](https://github.com/tensorflow/tensorflow/issues/4965) about HMC in tensorflow.

3. There are a few teams also [in progress](https://github.com/probtorch/pytorch/projects/1) to implement something similar in PyTorch. You can find their discussion of the design here: [https://goo.gl/9ccYsq](https://goo.gl/9ccYsq).

---

<div class="post-metadata">

**Author:** ![sarah\_s388](https://yyz2.discourse-cdn.com/flex036/user_avatar/discourse.pymc.io/sarah_s388/32/1221_2.png) [@sarah\_s388](https://discourse.pymc.io/u/sarah_s388)\
**Post date:** [October 30, 2018, 7:08am UTC](https://discourse.pymc.io/t/12-2017-tensorflow-distributions/769/2 "2018-10-30T07:08:41Z")

</div>

Hello Mate,

Thanks for sharing info on TensorFlow Distributions, I have been [learning tensorflow](https://goo.gl/Vn9gFb) for mindmajix.