Merge branch 'bump_v3.1rc4' of github.com:pymc-devs/pymc3 into bump_v3.1rc4

Author: fonnesbeck

.readthedocs.yml

@@ -3,7 +3,7 @@ formats:
   - epub
 
 # pip requirements file for building sphinx docs (e.g. nbsphinx)
-requirements_file: docs/requirements-docs.txt
+requirements_file: docs/requirements-dev.txt
 
 # can also specify a conda env file if pip isn't enough, for example:
 # conda:

RELEASE-NOTES.md

@@ -4,10 +4,26 @@
 
 ### New features
 
-* [Add Gaussian Process submodule](http://pymc-devs.github.io/pymc3/notebooks/GP-introduction.html)
+* New user forum at http://discourse.pymc.io
+
+* [Gaussian Process submodule](http://pymc-devs.github.io/pymc3/notebooks/GP-introduction.html)
+
+* Much improved variational inference support:
+
+  - [Add Operator Variational Inference (experimental).](http://pymc-devs.github.io/pymc3/notebooks/bayesian_neural_network_opvi-advi.html)
+
+  - [Add Stein-Variational Gradient Descent as well as Amortized SVGD (experimental).](https://github.com/pymc-devs/pymc3/pull/2183)
+
+  - [Add pm.Minibatch() to easily specify mini-batches.](http://pymc-devs.github.io/pymc3/notebooks/bayesian_neural_network_opvi-advi.html#Minibatch-ADVI)
+
+  - Added various optimizers including ADAM.
+  
+  - Stopping criterion implemented via callbacks.
 
 * sample() defaults changed: tuning is enabled for the first 500 samples which are then discarded from the trace as burn-in.
 
+* MvNormal supports Cholesky Decomposition now for increased speed and numerical stability.
+
 * Many optimizations and speed-ups.
 
 * NUTS implementation now matches current Stan implementation.
@@ -54,8 +70,6 @@
 
 * GLM models do not need a left-hand variable anymore.
 
-* Add support for cholesky parametrizations for cov/corr matrices.
-
 * Refactored HMC and NUTS for better readability.
 
 * Add support for Python 3.6.